thesis.lyx 337 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722472347244725472647274728472947304731473247334734473547364737473847394740474147424743474447454746474747484749475047514752475347544755475647574758475947604761476247634764476547664767476847694770477147724773477447754776477747784779478047814782478347844785478647874788478947904791479247934794479547964797479847994800480148024803480448054806480748084809481048114812481348144815481648174818481948204821482248234824482548264827482848294830483148324833483448354836483748384839484048414842484348444845484648474848484948504851485248534854485548564857485848594860486148624863486448654866486748684869487048714872487348744875487648774878487948804881488248834884488548864887488848894890489148924893489448954896489748984899490049014902490349044905490649074908490949104911491249134914491549164917491849194920492149224923492449254926492749284929493049314932493349344935493649374938493949404941494249434944494549464947494849494950495149524953495449554956495749584959496049614962496349644965496649674968496949704971497249734974497549764977497849794980498149824983498449854986498749884989499049914992499349944995499649974998499950005001500250035004500550065007500850095010501150125013501450155016501750185019502050215022502350245025502650275028502950305031503250335034503550365037503850395040504150425043504450455046504750485049505050515052505350545055505650575058505950605061506250635064506550665067506850695070507150725073507450755076507750785079508050815082508350845085508650875088508950905091509250935094509550965097509850995100510151025103510451055106510751085109511051115112511351145115511651175118511951205121512251235124512551265127512851295130513151325133513451355136513751385139514051415142514351445145514651475148514951505151515251535154515551565157515851595160516151625163516451655166516751685169517051715172517351745175517651775178517951805181518251835184518551865187518851895190519151925193519451955196519751985199520052015202520352045205520652075208520952105211521252135214521552165217521852195220522152225223522452255226522752285229523052315232523352345235523652375238523952405241524252435244524552465247524852495250525152525253525452555256525752585259526052615262526352645265526652675268526952705271527252735274527552765277527852795280528152825283528452855286528752885289529052915292529352945295529652975298529953005301530253035304530553065307530853095310531153125313531453155316531753185319532053215322532353245325532653275328532953305331533253335334533553365337533853395340534153425343534453455346534753485349535053515352535353545355535653575358535953605361536253635364536553665367536853695370537153725373537453755376537753785379538053815382538353845385538653875388538953905391539253935394539553965397539853995400540154025403540454055406540754085409541054115412541354145415541654175418541954205421542254235424542554265427542854295430543154325433543454355436543754385439544054415442544354445445544654475448544954505451545254535454545554565457545854595460546154625463546454655466546754685469547054715472547354745475547654775478547954805481548254835484548554865487548854895490549154925493549454955496549754985499550055015502550355045505550655075508550955105511551255135514551555165517551855195520552155225523552455255526552755285529553055315532553355345535553655375538553955405541554255435544554555465547554855495550555155525553555455555556555755585559556055615562556355645565556655675568556955705571557255735574557555765577557855795580558155825583558455855586558755885589559055915592559355945595559655975598559956005601560256035604560556065607560856095610561156125613561456155616561756185619562056215622562356245625562656275628562956305631563256335634563556365637563856395640564156425643564456455646564756485649565056515652565356545655565656575658565956605661566256635664566556665667566856695670567156725673567456755676567756785679568056815682568356845685568656875688568956905691569256935694569556965697569856995700570157025703570457055706570757085709571057115712571357145715571657175718571957205721572257235724572557265727572857295730573157325733573457355736573757385739574057415742574357445745574657475748574957505751575257535754575557565757575857595760576157625763576457655766576757685769577057715772577357745775577657775778577957805781578257835784578557865787578857895790579157925793579457955796579757985799580058015802580358045805580658075808580958105811581258135814581558165817581858195820582158225823582458255826582758285829583058315832583358345835583658375838583958405841584258435844584558465847584858495850585158525853585458555856585758585859586058615862586358645865586658675868586958705871587258735874587558765877587858795880588158825883588458855886588758885889589058915892589358945895589658975898589959005901590259035904590559065907590859095910591159125913591459155916591759185919592059215922592359245925592659275928592959305931593259335934593559365937593859395940594159425943594459455946594759485949595059515952595359545955595659575958595959605961596259635964596559665967596859695970597159725973597459755976597759785979598059815982598359845985598659875988598959905991599259935994599559965997599859996000600160026003600460056006600760086009601060116012601360146015601660176018601960206021602260236024602560266027602860296030603160326033603460356036603760386039604060416042604360446045604660476048604960506051605260536054605560566057605860596060606160626063606460656066606760686069607060716072607360746075607660776078607960806081608260836084608560866087608860896090609160926093609460956096609760986099610061016102610361046105610661076108610961106111611261136114611561166117611861196120612161226123612461256126612761286129613061316132613361346135613661376138613961406141614261436144614561466147614861496150615161526153615461556156615761586159616061616162616361646165616661676168616961706171617261736174617561766177617861796180618161826183618461856186618761886189619061916192619361946195619661976198619962006201620262036204620562066207620862096210621162126213621462156216621762186219622062216222622362246225622662276228622962306231623262336234623562366237623862396240624162426243624462456246624762486249625062516252625362546255625662576258625962606261626262636264626562666267626862696270627162726273627462756276627762786279628062816282628362846285628662876288628962906291629262936294629562966297629862996300630163026303630463056306630763086309631063116312631363146315631663176318631963206321632263236324632563266327632863296330633163326333633463356336633763386339634063416342634363446345634663476348634963506351635263536354635563566357635863596360636163626363636463656366636763686369637063716372637363746375637663776378637963806381638263836384638563866387638863896390639163926393639463956396639763986399640064016402640364046405640664076408640964106411641264136414641564166417641864196420642164226423642464256426642764286429643064316432643364346435643664376438643964406441644264436444644564466447644864496450645164526453645464556456645764586459646064616462646364646465646664676468646964706471647264736474647564766477647864796480648164826483648464856486648764886489649064916492649364946495649664976498649965006501650265036504650565066507650865096510651165126513651465156516651765186519652065216522652365246525652665276528652965306531653265336534653565366537653865396540654165426543654465456546654765486549655065516552655365546555655665576558655965606561656265636564656565666567656865696570657165726573657465756576657765786579658065816582658365846585658665876588658965906591659265936594659565966597659865996600660166026603660466056606660766086609661066116612661366146615661666176618661966206621662266236624662566266627662866296630663166326633663466356636663766386639664066416642664366446645664666476648664966506651665266536654665566566657665866596660666166626663666466656666666766686669667066716672667366746675667666776678667966806681668266836684668566866687668866896690669166926693669466956696669766986699670067016702670367046705670667076708670967106711671267136714671567166717671867196720672167226723672467256726672767286729673067316732673367346735673667376738673967406741674267436744674567466747674867496750675167526753675467556756675767586759676067616762676367646765676667676768676967706771677267736774677567766777677867796780678167826783678467856786678767886789679067916792679367946795679667976798679968006801680268036804680568066807680868096810681168126813681468156816681768186819682068216822682368246825682668276828682968306831683268336834683568366837683868396840684168426843684468456846684768486849685068516852685368546855685668576858685968606861686268636864686568666867686868696870687168726873687468756876687768786879688068816882688368846885688668876888688968906891689268936894689568966897689868996900690169026903690469056906690769086909691069116912691369146915691669176918691969206921692269236924692569266927692869296930693169326933693469356936693769386939694069416942694369446945694669476948694969506951695269536954695569566957695869596960696169626963696469656966696769686969697069716972697369746975697669776978697969806981698269836984698569866987698869896990699169926993699469956996699769986999700070017002700370047005700670077008700970107011701270137014701570167017701870197020702170227023702470257026702770287029703070317032703370347035703670377038703970407041704270437044704570467047704870497050705170527053705470557056705770587059706070617062706370647065706670677068706970707071707270737074707570767077707870797080708170827083708470857086708770887089709070917092709370947095709670977098709971007101710271037104710571067107710871097110711171127113711471157116711771187119712071217122712371247125712671277128712971307131713271337134713571367137713871397140714171427143714471457146714771487149715071517152715371547155715671577158715971607161716271637164716571667167716871697170717171727173717471757176717771787179718071817182718371847185718671877188718971907191719271937194719571967197719871997200720172027203720472057206720772087209721072117212721372147215721672177218721972207221722272237224722572267227722872297230723172327233723472357236723772387239724072417242724372447245724672477248724972507251725272537254725572567257725872597260726172627263726472657266726772687269727072717272727372747275727672777278727972807281728272837284728572867287728872897290729172927293729472957296729772987299730073017302730373047305730673077308730973107311731273137314731573167317731873197320732173227323732473257326732773287329733073317332733373347335733673377338733973407341734273437344734573467347734873497350735173527353735473557356735773587359736073617362736373647365736673677368736973707371737273737374737573767377737873797380738173827383738473857386738773887389739073917392739373947395739673977398739974007401740274037404740574067407740874097410741174127413741474157416741774187419742074217422742374247425742674277428742974307431743274337434743574367437743874397440744174427443744474457446744774487449745074517452745374547455745674577458745974607461746274637464746574667467746874697470747174727473747474757476747774787479748074817482748374847485748674877488748974907491749274937494749574967497749874997500750175027503750475057506750775087509751075117512751375147515751675177518751975207521752275237524752575267527752875297530753175327533753475357536753775387539754075417542754375447545754675477548754975507551755275537554755575567557755875597560756175627563756475657566756775687569757075717572757375747575757675777578757975807581758275837584758575867587758875897590759175927593759475957596759775987599760076017602760376047605760676077608760976107611761276137614761576167617761876197620762176227623762476257626762776287629763076317632763376347635763676377638763976407641764276437644764576467647764876497650765176527653765476557656765776587659766076617662766376647665766676677668766976707671767276737674767576767677767876797680768176827683768476857686768776887689769076917692769376947695769676977698769977007701770277037704770577067707770877097710771177127713771477157716771777187719772077217722772377247725772677277728772977307731773277337734773577367737773877397740774177427743774477457746774777487749775077517752775377547755775677577758775977607761776277637764776577667767776877697770777177727773777477757776777777787779778077817782778377847785778677877788778977907791779277937794779577967797779877997800780178027803780478057806780778087809781078117812781378147815781678177818781978207821782278237824782578267827782878297830783178327833783478357836783778387839784078417842784378447845784678477848784978507851785278537854785578567857785878597860786178627863786478657866786778687869787078717872787378747875787678777878787978807881788278837884788578867887788878897890789178927893789478957896789778987899790079017902790379047905790679077908790979107911791279137914791579167917791879197920792179227923792479257926792779287929793079317932793379347935793679377938793979407941794279437944794579467947794879497950795179527953795479557956795779587959796079617962796379647965796679677968796979707971797279737974797579767977797879797980798179827983798479857986798779887989799079917992799379947995799679977998799980008001800280038004800580068007800880098010801180128013801480158016801780188019802080218022802380248025802680278028802980308031803280338034803580368037803880398040804180428043804480458046804780488049805080518052805380548055805680578058805980608061806280638064806580668067806880698070807180728073807480758076807780788079808080818082808380848085808680878088808980908091809280938094809580968097809880998100810181028103810481058106810781088109811081118112811381148115811681178118811981208121812281238124812581268127812881298130813181328133813481358136813781388139814081418142814381448145814681478148814981508151815281538154815581568157815881598160816181628163816481658166816781688169817081718172817381748175817681778178817981808181818281838184818581868187818881898190819181928193819481958196819781988199820082018202820382048205820682078208820982108211821282138214821582168217821882198220822182228223822482258226822782288229823082318232823382348235823682378238823982408241824282438244824582468247824882498250825182528253825482558256825782588259826082618262826382648265826682678268826982708271827282738274827582768277827882798280828182828283828482858286828782888289829082918292829382948295829682978298829983008301830283038304830583068307830883098310831183128313831483158316831783188319832083218322832383248325832683278328832983308331833283338334833583368337833883398340834183428343834483458346834783488349835083518352835383548355835683578358835983608361836283638364836583668367836883698370837183728373837483758376837783788379838083818382838383848385838683878388838983908391839283938394839583968397839883998400840184028403840484058406840784088409841084118412841384148415841684178418841984208421842284238424842584268427842884298430843184328433843484358436843784388439844084418442844384448445844684478448844984508451845284538454845584568457845884598460846184628463846484658466846784688469847084718472847384748475847684778478847984808481848284838484848584868487848884898490849184928493849484958496849784988499850085018502850385048505850685078508850985108511851285138514851585168517851885198520852185228523852485258526852785288529853085318532853385348535853685378538853985408541854285438544854585468547854885498550855185528553855485558556855785588559856085618562856385648565856685678568856985708571857285738574857585768577857885798580858185828583858485858586858785888589859085918592859385948595859685978598859986008601860286038604860586068607860886098610861186128613861486158616861786188619862086218622862386248625862686278628862986308631863286338634863586368637863886398640864186428643864486458646864786488649865086518652865386548655865686578658865986608661866286638664866586668667866886698670867186728673867486758676867786788679868086818682868386848685868686878688868986908691869286938694869586968697869886998700870187028703870487058706870787088709871087118712871387148715871687178718871987208721872287238724872587268727872887298730873187328733873487358736873787388739874087418742874387448745874687478748874987508751875287538754875587568757875887598760876187628763876487658766876787688769877087718772877387748775877687778778877987808781878287838784878587868787878887898790879187928793879487958796879787988799880088018802880388048805880688078808880988108811881288138814881588168817881888198820882188228823882488258826882788288829883088318832883388348835883688378838883988408841884288438844884588468847884888498850885188528853885488558856885788588859886088618862886388648865886688678868886988708871887288738874887588768877887888798880888188828883888488858886888788888889889088918892889388948895889688978898889989008901890289038904890589068907890889098910891189128913891489158916891789188919892089218922892389248925892689278928892989308931893289338934893589368937893889398940894189428943894489458946894789488949895089518952895389548955895689578958895989608961896289638964896589668967896889698970897189728973897489758976897789788979898089818982898389848985898689878988898989908991899289938994899589968997899889999000900190029003900490059006900790089009901090119012901390149015901690179018901990209021902290239024902590269027902890299030903190329033903490359036903790389039904090419042904390449045904690479048904990509051905290539054905590569057905890599060906190629063906490659066906790689069907090719072907390749075907690779078907990809081908290839084908590869087908890899090909190929093909490959096909790989099910091019102910391049105910691079108910991109111911291139114911591169117911891199120912191229123912491259126912791289129913091319132913391349135913691379138913991409141914291439144914591469147914891499150915191529153915491559156915791589159916091619162916391649165916691679168916991709171917291739174917591769177917891799180918191829183918491859186918791889189919091919192919391949195919691979198919992009201920292039204920592069207920892099210921192129213921492159216921792189219922092219222922392249225922692279228922992309231923292339234923592369237923892399240924192429243924492459246924792489249925092519252925392549255925692579258925992609261926292639264926592669267926892699270927192729273927492759276927792789279928092819282928392849285928692879288928992909291929292939294929592969297929892999300930193029303930493059306930793089309931093119312931393149315931693179318931993209321932293239324932593269327932893299330933193329333933493359336933793389339934093419342934393449345934693479348934993509351935293539354935593569357935893599360936193629363936493659366936793689369937093719372937393749375937693779378937993809381938293839384938593869387938893899390939193929393939493959396939793989399940094019402940394049405940694079408940994109411941294139414941594169417941894199420942194229423942494259426942794289429943094319432943394349435943694379438943994409441944294439444944594469447944894499450945194529453945494559456945794589459946094619462946394649465946694679468946994709471947294739474947594769477947894799480948194829483948494859486948794889489949094919492949394949495949694979498949995009501950295039504950595069507950895099510951195129513951495159516951795189519952095219522952395249525952695279528952995309531953295339534953595369537953895399540954195429543954495459546954795489549955095519552955395549555955695579558955995609561956295639564956595669567956895699570957195729573957495759576957795789579958095819582958395849585958695879588958995909591959295939594959595969597959895999600960196029603960496059606960796089609961096119612961396149615961696179618961996209621962296239624962596269627962896299630963196329633963496359636963796389639964096419642964396449645964696479648964996509651965296539654965596569657965896599660966196629663966496659666966796689669967096719672967396749675967696779678967996809681968296839684968596869687968896899690969196929693969496959696969796989699970097019702970397049705970697079708970997109711971297139714971597169717971897199720972197229723972497259726972797289729973097319732973397349735973697379738973997409741974297439744974597469747974897499750975197529753975497559756975797589759976097619762976397649765976697679768976997709771977297739774977597769777977897799780978197829783978497859786978797889789979097919792979397949795979697979798979998009801980298039804980598069807980898099810981198129813981498159816981798189819982098219822982398249825982698279828982998309831983298339834983598369837983898399840984198429843984498459846984798489849985098519852985398549855985698579858985998609861986298639864986598669867986898699870987198729873987498759876987798789879988098819882988398849885988698879888988998909891989298939894989598969897989898999900990199029903990499059906990799089909991099119912991399149915991699179918991999209921992299239924992599269927992899299930993199329933993499359936993799389939994099419942994399449945994699479948994999509951995299539954995599569957995899599960996199629963996499659966996799689969997099719972997399749975997699779978997999809981998299839984998599869987998899899990999199929993999499959996999799989999100001000110002100031000410005100061000710008100091001010011100121001310014100151001610017100181001910020100211002210023100241002510026100271002810029100301003110032100331003410035100361003710038100391004010041100421004310044100451004610047100481004910050100511005210053100541005510056100571005810059100601006110062100631006410065100661006710068100691007010071100721007310074100751007610077100781007910080100811008210083100841008510086100871008810089100901009110092100931009410095100961009710098100991010010101101021010310104101051010610107101081010910110101111011210113101141011510116101171011810119101201012110122101231012410125101261012710128101291013010131101321013310134101351013610137101381013910140101411014210143101441014510146101471014810149101501015110152101531015410155101561015710158101591016010161101621016310164101651016610167101681016910170101711017210173101741017510176101771017810179101801018110182101831018410185101861018710188101891019010191101921019310194101951019610197101981019910200102011020210203102041020510206102071020810209102101021110212102131021410215102161021710218102191022010221102221022310224102251022610227102281022910230102311023210233102341023510236102371023810239102401024110242102431024410245102461024710248102491025010251102521025310254102551025610257102581025910260102611026210263102641026510266102671026810269102701027110272102731027410275102761027710278102791028010281102821028310284102851028610287102881028910290102911029210293102941029510296102971029810299103001030110302103031030410305103061030710308103091031010311103121031310314103151031610317103181031910320103211032210323103241032510326103271032810329103301033110332103331033410335103361033710338103391034010341103421034310344103451034610347103481034910350103511035210353103541035510356103571035810359103601036110362103631036410365103661036710368103691037010371103721037310374103751037610377103781037910380103811038210383103841038510386103871038810389103901039110392103931039410395103961039710398103991040010401104021040310404104051040610407104081040910410104111041210413104141041510416104171041810419104201042110422104231042410425104261042710428104291043010431104321043310434104351043610437104381043910440104411044210443104441044510446104471044810449104501045110452104531045410455104561045710458104591046010461104621046310464104651046610467104681046910470104711047210473104741047510476104771047810479104801048110482104831048410485104861048710488104891049010491104921049310494104951049610497104981049910500105011050210503105041050510506105071050810509105101051110512105131051410515105161051710518105191052010521105221052310524105251052610527105281052910530105311053210533105341053510536105371053810539105401054110542105431054410545105461054710548105491055010551105521055310554105551055610557105581055910560105611056210563105641056510566105671056810569105701057110572105731057410575105761057710578105791058010581105821058310584105851058610587105881058910590105911059210593105941059510596105971059810599106001060110602106031060410605106061060710608106091061010611106121061310614106151061610617106181061910620106211062210623106241062510626106271062810629106301063110632106331063410635106361063710638106391064010641106421064310644106451064610647106481064910650106511065210653106541065510656106571065810659106601066110662106631066410665106661066710668106691067010671106721067310674106751067610677106781067910680106811068210683106841068510686106871068810689106901069110692106931069410695106961069710698106991070010701107021070310704107051070610707107081070910710107111071210713107141071510716107171071810719107201072110722107231072410725107261072710728107291073010731107321073310734107351073610737107381073910740107411074210743107441074510746107471074810749107501075110752107531075410755107561075710758107591076010761107621076310764107651076610767107681076910770107711077210773107741077510776107771077810779107801078110782107831078410785107861078710788107891079010791107921079310794107951079610797107981079910800108011080210803108041080510806108071080810809108101081110812108131081410815108161081710818108191082010821108221082310824108251082610827108281082910830108311083210833108341083510836108371083810839108401084110842108431084410845108461084710848108491085010851108521085310854108551085610857108581085910860108611086210863108641086510866108671086810869108701087110872108731087410875108761087710878108791088010881108821088310884108851088610887108881088910890108911089210893108941089510896108971089810899109001090110902109031090410905109061090710908109091091010911109121091310914109151091610917109181091910920109211092210923109241092510926109271092810929109301093110932109331093410935109361093710938109391094010941109421094310944109451094610947109481094910950109511095210953109541095510956109571095810959109601096110962109631096410965109661096710968109691097010971109721097310974109751097610977109781097910980109811098210983109841098510986109871098810989109901099110992109931099410995109961099710998109991100011001110021100311004110051100611007110081100911010110111101211013110141101511016110171101811019110201102111022110231102411025110261102711028110291103011031110321103311034110351103611037110381103911040110411104211043110441104511046110471104811049110501105111052110531105411055110561105711058110591106011061110621106311064110651106611067110681106911070110711107211073110741107511076110771107811079110801108111082110831108411085110861108711088110891109011091110921109311094110951109611097110981109911100111011110211103111041110511106111071110811109111101111111112111131111411115111161111711118111191112011121111221112311124111251112611127111281112911130111311113211133111341113511136111371113811139111401114111142111431114411145111461114711148111491115011151111521115311154111551115611157111581115911160111611116211163111641116511166111671116811169111701117111172111731117411175111761117711178111791118011181111821118311184111851118611187111881118911190111911119211193111941119511196111971119811199112001120111202112031120411205112061120711208112091121011211112121121311214112151121611217112181121911220112211122211223112241122511226112271122811229112301123111232112331123411235112361123711238112391124011241112421124311244112451124611247112481124911250112511125211253112541125511256112571125811259112601126111262112631126411265112661126711268112691127011271112721127311274112751127611277112781127911280112811128211283112841128511286112871128811289112901129111292112931129411295112961129711298112991130011301113021130311304113051130611307113081130911310113111131211313113141131511316113171131811319113201132111322113231132411325113261132711328113291133011331113321133311334113351133611337113381133911340113411134211343113441134511346113471134811349113501135111352113531135411355113561135711358113591136011361113621136311364113651136611367113681136911370113711137211373113741137511376113771137811379113801138111382113831138411385113861138711388113891139011391113921139311394113951139611397113981139911400114011140211403114041140511406114071140811409114101141111412114131141411415114161141711418114191142011421114221142311424114251142611427114281142911430114311143211433114341143511436114371143811439114401144111442114431144411445114461144711448114491145011451114521145311454114551145611457114581145911460114611146211463114641146511466114671146811469114701147111472114731147411475114761147711478114791148011481114821148311484114851148611487114881148911490114911149211493114941149511496114971149811499115001150111502115031150411505115061150711508115091151011511115121151311514115151151611517115181151911520115211152211523115241152511526115271152811529115301153111532115331153411535115361153711538115391154011541115421154311544115451154611547115481154911550115511155211553115541155511556115571155811559115601156111562115631156411565115661156711568115691157011571115721157311574115751157611577115781157911580115811158211583115841158511586115871158811589115901159111592115931159411595115961159711598115991160011601116021160311604116051160611607116081160911610116111161211613116141161511616116171161811619116201162111622116231162411625116261162711628116291163011631116321163311634116351163611637116381163911640116411164211643116441164511646116471164811649116501165111652116531165411655116561165711658116591166011661116621166311664116651166611667116681166911670116711167211673116741167511676116771167811679116801168111682116831168411685116861168711688116891169011691116921169311694116951169611697116981169911700117011170211703117041170511706117071170811709117101171111712117131171411715117161171711718117191172011721117221172311724117251172611727117281172911730117311173211733117341173511736117371173811739117401174111742117431174411745117461174711748117491175011751117521175311754117551175611757117581175911760117611176211763117641176511766117671176811769117701177111772117731177411775117761177711778117791178011781117821178311784117851178611787117881178911790117911179211793117941179511796117971179811799118001180111802118031180411805118061180711808118091181011811118121181311814118151181611817118181181911820118211182211823118241182511826118271182811829118301183111832118331183411835118361183711838118391184011841118421184311844118451184611847118481184911850118511185211853118541185511856118571185811859118601186111862118631186411865118661186711868118691187011871118721187311874118751187611877118781187911880118811188211883118841188511886118871188811889118901189111892118931189411895118961189711898118991190011901119021190311904119051190611907119081190911910119111191211913119141191511916119171191811919119201192111922119231192411925119261192711928119291193011931119321193311934119351193611937119381193911940119411194211943119441194511946119471194811949119501195111952119531195411955119561195711958119591196011961119621196311964119651196611967119681196911970119711197211973119741197511976119771197811979119801198111982119831198411985119861198711988119891199011991119921199311994119951199611997119981199912000120011200212003120041200512006120071200812009120101201112012120131201412015120161201712018120191202012021120221202312024120251202612027120281202912030120311203212033120341203512036120371203812039120401204112042120431204412045120461204712048120491205012051120521205312054120551205612057120581205912060120611206212063120641206512066120671206812069120701207112072120731207412075120761207712078120791208012081120821208312084120851208612087120881208912090120911209212093120941209512096120971209812099121001210112102121031210412105121061210712108121091211012111121121211312114121151211612117121181211912120121211212212123121241212512126121271212812129121301213112132121331213412135121361213712138121391214012141121421214312144121451214612147121481214912150121511215212153121541215512156121571215812159121601216112162121631216412165121661216712168121691217012171121721217312174121751217612177121781217912180121811218212183121841218512186121871218812189121901219112192121931219412195121961219712198121991220012201122021220312204122051220612207122081220912210122111221212213122141221512216122171221812219122201222112222122231222412225122261222712228122291223012231122321223312234122351223612237122381223912240122411224212243122441224512246122471224812249122501225112252122531225412255122561225712258122591226012261122621226312264122651226612267122681226912270122711227212273122741227512276122771227812279122801228112282122831228412285122861228712288122891229012291122921229312294122951229612297122981229912300123011230212303123041230512306123071230812309123101231112312123131231412315123161231712318123191232012321123221232312324123251232612327123281232912330123311233212333123341233512336123371233812339123401234112342123431234412345123461234712348123491235012351123521235312354123551235612357123581235912360123611236212363123641236512366123671236812369123701237112372123731237412375123761237712378123791238012381123821238312384123851238612387123881238912390123911239212393123941239512396123971239812399124001240112402124031240412405124061240712408124091241012411124121241312414124151241612417124181241912420124211242212423124241242512426124271242812429124301243112432124331243412435124361243712438124391244012441124421244312444124451244612447124481244912450124511245212453124541245512456124571245812459124601246112462124631246412465124661246712468124691247012471124721247312474124751247612477124781247912480124811248212483124841248512486124871248812489124901249112492124931249412495124961249712498124991250012501125021250312504125051250612507125081250912510125111251212513125141251512516125171251812519125201252112522125231252412525125261252712528125291253012531125321253312534125351253612537125381253912540125411254212543125441254512546125471254812549125501255112552125531255412555125561255712558125591256012561125621256312564125651256612567125681256912570125711257212573125741257512576125771257812579125801258112582125831258412585125861258712588125891259012591125921259312594125951259612597125981259912600126011260212603126041260512606126071260812609126101261112612126131261412615126161261712618126191262012621126221262312624126251262612627126281262912630126311263212633126341263512636126371263812639126401264112642126431264412645126461264712648126491265012651126521265312654126551265612657126581265912660126611266212663126641266512666126671266812669126701267112672126731267412675126761267712678126791268012681126821268312684126851268612687126881268912690126911269212693126941269512696126971269812699127001270112702127031270412705127061270712708127091271012711127121271312714127151271612717127181271912720127211272212723127241272512726127271272812729127301273112732127331273412735127361273712738127391274012741127421274312744127451274612747127481274912750127511275212753127541275512756127571275812759127601276112762127631276412765127661276712768127691277012771127721277312774127751277612777127781277912780127811278212783127841278512786127871278812789127901279112792127931279412795127961279712798127991280012801128021280312804128051280612807128081280912810128111281212813128141281512816128171281812819128201282112822128231282412825128261282712828128291283012831128321283312834128351283612837128381283912840128411284212843128441284512846128471284812849128501285112852128531285412855128561285712858128591286012861128621286312864128651286612867128681286912870128711287212873128741287512876128771287812879128801288112882128831288412885128861288712888128891289012891128921289312894128951289612897128981289912900129011290212903129041290512906129071290812909129101291112912129131291412915129161291712918129191292012921129221292312924129251292612927129281292912930129311293212933129341293512936129371293812939129401294112942129431294412945129461294712948129491295012951129521295312954129551295612957129581295912960129611296212963129641296512966129671296812969129701297112972129731297412975129761297712978129791298012981129821298312984129851298612987129881298912990129911299212993129941299512996129971299812999130001300113002130031300413005130061300713008130091301013011130121301313014130151301613017130181301913020130211302213023130241302513026130271302813029130301303113032130331303413035130361303713038130391304013041130421304313044130451304613047130481304913050130511305213053130541305513056130571305813059130601306113062130631306413065130661306713068130691307013071130721307313074130751307613077130781307913080130811308213083130841308513086130871308813089130901309113092130931309413095130961309713098130991310013101131021310313104131051310613107131081310913110131111311213113131141311513116131171311813119131201312113122131231312413125131261312713128131291313013131131321313313134131351313613137131381313913140131411314213143131441314513146131471314813149131501315113152131531315413155131561315713158131591316013161131621316313164131651316613167131681316913170131711317213173131741317513176131771317813179131801318113182131831318413185131861318713188131891319013191131921319313194131951319613197131981319913200132011320213203132041320513206132071320813209132101321113212132131321413215132161321713218132191322013221132221322313224132251322613227132281322913230132311323213233132341323513236132371323813239132401324113242132431324413245132461324713248132491325013251132521325313254132551325613257132581325913260132611326213263132641326513266132671326813269132701327113272132731327413275132761327713278132791328013281132821328313284132851328613287132881328913290132911329213293132941329513296132971329813299133001330113302133031330413305133061330713308133091331013311133121331313314133151331613317133181331913320133211332213323133241332513326133271332813329133301333113332133331333413335133361333713338133391334013341133421334313344133451334613347133481334913350133511335213353133541335513356133571335813359133601336113362133631336413365133661336713368133691337013371133721337313374133751337613377133781337913380133811338213383133841338513386133871338813389133901339113392133931339413395133961339713398133991340013401134021340313404134051340613407134081340913410134111341213413134141341513416134171341813419134201342113422134231342413425134261342713428134291343013431134321343313434134351343613437134381343913440134411344213443134441344513446134471344813449134501345113452134531345413455134561345713458134591346013461134621346313464134651346613467134681346913470134711347213473134741347513476134771347813479134801348113482134831348413485134861348713488134891349013491134921349313494134951349613497134981349913500135011350213503135041350513506135071350813509135101351113512135131351413515135161351713518135191352013521135221352313524135251352613527135281352913530135311353213533135341353513536135371353813539135401354113542135431354413545135461354713548135491355013551135521355313554135551355613557135581355913560135611356213563135641356513566135671356813569135701357113572135731357413575135761357713578135791358013581135821358313584135851358613587135881358913590135911359213593135941359513596135971359813599136001360113602136031360413605136061360713608136091361013611136121361313614136151361613617136181361913620136211362213623136241362513626136271362813629136301363113632136331363413635136361363713638136391364013641136421364313644136451364613647136481364913650136511365213653136541365513656136571365813659136601366113662136631366413665136661366713668136691367013671136721367313674136751367613677136781367913680136811368213683136841368513686136871368813689136901369113692136931369413695136961369713698136991370013701137021370313704137051370613707137081370913710137111371213713137141371513716137171371813719137201372113722137231372413725137261372713728137291373013731137321373313734137351373613737137381373913740137411374213743137441374513746137471374813749137501375113752137531375413755137561375713758137591376013761137621376313764137651376613767137681376913770137711377213773137741377513776137771377813779137801378113782137831378413785137861378713788137891379013791137921379313794137951379613797137981379913800138011380213803138041380513806138071380813809138101381113812138131381413815138161381713818138191382013821138221382313824138251382613827138281382913830138311383213833138341383513836138371383813839138401384113842138431384413845138461384713848138491385013851138521385313854138551385613857138581385913860138611386213863138641386513866138671386813869138701387113872138731387413875138761387713878138791388013881138821388313884138851388613887138881388913890138911389213893138941389513896138971389813899139001390113902139031390413905139061390713908139091391013911139121391313914139151391613917139181391913920139211392213923139241392513926139271392813929139301393113932139331393413935139361393713938139391394013941139421394313944139451394613947139481394913950139511395213953139541395513956139571395813959139601396113962139631396413965139661396713968139691397013971139721397313974139751397613977139781397913980139811398213983139841398513986139871398813989139901399113992139931399413995139961399713998139991400014001140021400314004140051400614007140081400914010140111401214013140141401514016140171401814019140201402114022140231402414025140261402714028140291403014031140321403314034140351403614037140381403914040140411404214043140441404514046140471404814049140501405114052140531405414055140561405714058140591406014061140621406314064140651406614067140681406914070140711407214073140741407514076140771407814079140801408114082140831408414085140861408714088140891409014091140921409314094140951409614097140981409914100141011410214103141041410514106141071410814109141101411114112141131411414115141161411714118141191412014121141221412314124141251412614127141281412914130141311413214133141341413514136141371413814139141401414114142141431414414145141461414714148141491415014151141521415314154141551415614157141581415914160141611416214163141641416514166141671416814169141701417114172141731417414175141761417714178141791418014181141821418314184141851418614187141881418914190141911419214193141941419514196141971419814199142001420114202142031420414205142061420714208142091421014211142121421314214142151421614217142181421914220142211422214223142241422514226142271422814229142301423114232142331423414235142361423714238142391424014241142421424314244142451424614247142481424914250142511425214253142541425514256142571425814259142601426114262142631426414265142661426714268142691427014271142721427314274142751427614277142781427914280142811428214283142841428514286142871428814289142901429114292142931429414295142961429714298142991430014301143021430314304143051430614307143081430914310143111431214313143141431514316143171431814319143201432114322143231432414325143261432714328143291433014331143321433314334143351433614337143381433914340143411434214343143441434514346143471434814349143501435114352143531435414355143561435714358143591436014361143621436314364143651436614367143681436914370143711437214373143741437514376143771437814379143801438114382143831438414385143861438714388143891439014391143921439314394143951439614397143981439914400144011440214403144041440514406144071440814409144101441114412144131441414415144161441714418144191442014421144221442314424144251442614427144281442914430144311443214433144341443514436144371443814439144401444114442144431444414445144461444714448144491445014451144521445314454144551445614457144581445914460144611446214463144641446514466144671446814469144701447114472144731447414475144761447714478144791448014481144821448314484144851448614487144881448914490144911449214493144941449514496144971449814499145001450114502145031450414505145061450714508145091451014511145121451314514145151451614517145181451914520145211452214523145241452514526145271452814529145301453114532145331453414535145361453714538145391454014541145421454314544145451454614547145481454914550145511455214553145541455514556145571455814559145601456114562145631456414565145661456714568145691457014571145721457314574145751457614577145781457914580145811458214583145841458514586145871458814589145901459114592145931459414595145961459714598145991460014601146021460314604146051460614607146081460914610146111461214613146141461514616146171461814619146201462114622146231462414625146261462714628146291463014631146321463314634146351463614637146381463914640146411464214643146441464514646146471464814649146501465114652146531465414655146561465714658146591466014661146621466314664146651466614667146681466914670146711467214673146741467514676146771467814679146801468114682146831468414685146861468714688146891469014691146921469314694146951469614697146981469914700147011470214703147041470514706147071470814709147101471114712147131471414715147161471714718147191472014721147221472314724147251472614727147281472914730147311473214733147341473514736147371473814739147401474114742147431474414745147461474714748147491475014751147521475314754147551475614757147581475914760147611476214763147641476514766147671476814769147701477114772147731477414775147761477714778147791478014781147821478314784147851478614787147881478914790147911479214793147941479514796147971479814799148001480114802148031480414805148061480714808148091481014811148121481314814148151481614817148181481914820148211482214823148241482514826148271482814829148301483114832148331483414835148361483714838148391484014841148421484314844148451484614847148481484914850148511485214853148541485514856148571485814859148601486114862148631486414865148661486714868148691487014871148721487314874148751487614877148781487914880148811488214883148841488514886148871488814889148901489114892148931489414895148961489714898148991490014901149021490314904149051490614907149081490914910149111491214913149141491514916149171491814919149201492114922149231492414925149261492714928149291493014931149321493314934149351493614937149381493914940149411494214943149441494514946149471494814949149501495114952149531495414955149561495714958149591496014961149621496314964149651496614967149681496914970149711497214973149741497514976149771497814979149801498114982149831498414985149861498714988149891499014991149921499314994149951499614997149981499915000150011500215003150041500515006150071500815009150101501115012150131501415015150161501715018150191502015021150221502315024150251502615027150281502915030150311503215033150341503515036150371503815039150401504115042150431504415045150461504715048150491505015051150521505315054150551505615057150581505915060150611506215063150641506515066150671506815069150701507115072150731507415075150761507715078150791508015081150821508315084150851508615087
  1. #LyX 2.3 created this file. For more info see http://www.lyx.org/
  2. \lyxformat 544
  3. \begin_document
  4. \begin_header
  5. \save_transient_properties true
  6. \origin unavailable
  7. \textclass extbook
  8. \begin_preamble
  9. % List all used files in log output
  10. \listfiles
  11. % Add a DRAFT watermark
  12. \usepackage{draftwatermark}
  13. \usepackage{accsupp}
  14. \SetWatermarkLightness{0.97}
  15. \SetWatermarkScale{1}
  16. % Make watermark not copyable (in Adobe Reader)
  17. \SetWatermarkText{\BeginAccSupp{method=escape,ActualText={}}DRAFT\EndAccSupp{}}
  18. % Set up required header format
  19. \usepackage{fancyhdr}
  20. \pagestyle{fancy}
  21. \renewcommand{\headrulewidth}{0pt}
  22. \rhead{}
  23. \lhead{}
  24. \chead{}
  25. \rfoot{}
  26. \lfoot{}
  27. % Make page number not copyable (in Adobe Reader)
  28. \cfoot{\BeginAccSupp{method=escape,ActualText={}}\thepage\EndAccSupp{}} % Page number bottom center
  29. % Allow FloatBarrier command
  30. \usepackage{placeins}
  31. % Allow landscape pages
  32. \usepackage{pdflscape}
  33. % Allow doing things after the end of the current page
  34. % (to avoid landscape figures breaking up text)
  35. \usepackage{afterpage}
  36. % This one breaks subfigs so it's disabled
  37. % https://tex.stackexchange.com/questions/65680/automatically-bold-first-sentence-of-a-floats-caption
  38. \end_preamble
  39. \use_default_options true
  40. \begin_modules
  41. todonotes
  42. \end_modules
  43. \maintain_unincluded_children false
  44. \language english
  45. \language_package default
  46. \inputencoding utf8
  47. \fontencoding default
  48. \font_roman "default" "default"
  49. \font_sans "default" "default"
  50. \font_typewriter "default" "default"
  51. \font_math "auto" "auto"
  52. \font_default_family default
  53. \use_non_tex_fonts false
  54. \font_sc false
  55. \font_osf false
  56. \font_sf_scale 100 100
  57. \font_tt_scale 100 100
  58. \use_microtype false
  59. \use_dash_ligatures true
  60. \graphics default
  61. \default_output_format pdf4
  62. \output_sync 0
  63. \bibtex_command biber
  64. \index_command default
  65. \paperfontsize 12
  66. \spacing double
  67. \use_hyperref true
  68. \pdf_bookmarks true
  69. \pdf_bookmarksnumbered false
  70. \pdf_bookmarksopen false
  71. \pdf_bookmarksopenlevel 1
  72. \pdf_breaklinks false
  73. \pdf_pdfborder false
  74. \pdf_colorlinks false
  75. \pdf_backref false
  76. \pdf_pdfusetitle true
  77. \papersize letterpaper
  78. \use_geometry true
  79. \use_package amsmath 1
  80. \use_package amssymb 1
  81. \use_package cancel 1
  82. \use_package esint 1
  83. \use_package mathdots 1
  84. \use_package mathtools 1
  85. \use_package mhchem 1
  86. \use_package stackrel 1
  87. \use_package stmaryrd 1
  88. \use_package undertilde 1
  89. \cite_engine biblatex
  90. \cite_engine_type authoryear
  91. \biblio_style plain
  92. \biblatex_bibstyle authoryear
  93. \biblatex_citestyle numeric
  94. \use_bibtopic false
  95. \use_indices false
  96. \paperorientation portrait
  97. \suppress_date false
  98. \justification true
  99. \use_refstyle 1
  100. \use_minted 0
  101. \index Index
  102. \shortcut idx
  103. \color #008000
  104. \end_index
  105. \leftmargin 1.5in
  106. \topmargin 1in
  107. \rightmargin 1in
  108. \bottommargin 1in
  109. \secnumdepth 3
  110. \tocdepth 3
  111. \paragraph_separation indent
  112. \paragraph_indentation default
  113. \is_math_indent 0
  114. \math_numbering_side default
  115. \quotes_style english
  116. \dynamic_quotes 0
  117. \papercolumns 1
  118. \papersides 1
  119. \paperpagestyle default
  120. \tracking_changes false
  121. \output_changes false
  122. \html_math_output 0
  123. \html_css_as_file 0
  124. \html_be_strict false
  125. \end_header
  126. \begin_body
  127. \begin_layout Title
  128. Bioinformatic analysis of complex, high-throughput genomic and epigenomic
  129. data in the context of immunology and transplant rejection
  130. \end_layout
  131. \begin_layout Author
  132. A thesis presented
  133. \begin_inset Newline newline
  134. \end_inset
  135. by
  136. \begin_inset Newline newline
  137. \end_inset
  138. Ryan C.
  139. Thompson
  140. \begin_inset Newline newline
  141. \end_inset
  142. to
  143. \begin_inset Newline newline
  144. \end_inset
  145. The Scripps Research Institute Graduate Program
  146. \begin_inset Newline newline
  147. \end_inset
  148. in partial fulfillment of the requirements for the degree of
  149. \begin_inset Newline newline
  150. \end_inset
  151. Doctor of Philosophy in the subject of Biology
  152. \begin_inset Newline newline
  153. \end_inset
  154. for
  155. \begin_inset Newline newline
  156. \end_inset
  157. The Scripps Research Institute
  158. \begin_inset Newline newline
  159. \end_inset
  160. La Jolla, California
  161. \end_layout
  162. \begin_layout Date
  163. October 2019
  164. \end_layout
  165. \begin_layout Standard
  166. [Copyright notice]
  167. \end_layout
  168. \begin_layout Standard
  169. [Thesis acceptance form]
  170. \end_layout
  171. \begin_layout Standard
  172. [Dedication]
  173. \end_layout
  174. \begin_layout Standard
  175. [Acknowledgements]
  176. \end_layout
  177. \begin_layout Standard
  178. \begin_inset CommandInset toc
  179. LatexCommand tableofcontents
  180. \end_inset
  181. \end_layout
  182. \begin_layout Standard
  183. \begin_inset FloatList table
  184. \end_inset
  185. \end_layout
  186. \begin_layout Standard
  187. \begin_inset FloatList figure
  188. \end_inset
  189. \end_layout
  190. \begin_layout Standard
  191. [List of Abbreviations]
  192. \end_layout
  193. \begin_layout List of TODOs
  194. \end_layout
  195. \begin_layout Standard
  196. \begin_inset Flex TODO Note (inline)
  197. status open
  198. \begin_layout Plain Layout
  199. Check all figures to make sure they fit on the page with their legends.
  200. \end_layout
  201. \end_inset
  202. \end_layout
  203. \begin_layout Standard
  204. \begin_inset Flex TODO Note (inline)
  205. status open
  206. \begin_layout Plain Layout
  207. Search and replace: naive -> naïve
  208. \end_layout
  209. \end_inset
  210. \end_layout
  211. \begin_layout Standard
  212. \begin_inset Flex TODO Note (inline)
  213. status open
  214. \begin_layout Plain Layout
  215. Look into auto-generated nomenclature list: https://wiki.lyx.org/Tips/Nomenclature.
  216. Otherwise, do a manual pass for all abbreviations at the end.
  217. Do nomenclature/abbreviations independently for each chapter.
  218. \end_layout
  219. \end_inset
  220. \end_layout
  221. \begin_layout Standard
  222. \begin_inset Flex TODO Note (inline)
  223. status open
  224. \begin_layout Plain Layout
  225. Make all descriptions consistent in terms of
  226. \begin_inset Quotes eld
  227. \end_inset
  228. we did X
  229. \begin_inset Quotes erd
  230. \end_inset
  231. vs
  232. \begin_inset Quotes eld
  233. \end_inset
  234. I did X
  235. \begin_inset Quotes erd
  236. \end_inset
  237. vs
  238. \begin_inset Quotes eld
  239. \end_inset
  240. X was done
  241. \begin_inset Quotes erd
  242. \end_inset
  243. .
  244. \end_layout
  245. \end_inset
  246. \end_layout
  247. \begin_layout Chapter*
  248. Abstract
  249. \end_layout
  250. \begin_layout Standard
  251. \begin_inset Note Note
  252. status open
  253. \begin_layout Plain Layout
  254. It is included as an integral part of the thesis and should immediately
  255. precede the introduction.
  256. \end_layout
  257. \begin_layout Plain Layout
  258. Preparing your Abstract.
  259. Your abstract (a succinct description of your work) is limited to 350 words.
  260. UMI will shorten it if they must; please do not exceed the limit.
  261. \end_layout
  262. \begin_layout Itemize
  263. Include pertinent place names, names of persons (in full), and other proper
  264. nouns.
  265. These are useful in automated retrieval.
  266. \end_layout
  267. \begin_layout Itemize
  268. Display symbols, as well as foreign words and phrases, clearly and accurately.
  269. Include transliterations for characters other than Roman and Greek letters
  270. and Arabic numerals.
  271. Include accents and diacritical marks.
  272. \end_layout
  273. \begin_layout Itemize
  274. Do not include graphs, charts, tables, or illustrations in your abstract.
  275. \end_layout
  276. \end_inset
  277. \end_layout
  278. \begin_layout Standard
  279. \begin_inset Flex TODO Note (inline)
  280. status open
  281. \begin_layout Plain Layout
  282. Obviously the abstract gets written last.
  283. \end_layout
  284. \end_inset
  285. \end_layout
  286. \begin_layout Chapter*
  287. Notes to draft readers
  288. \end_layout
  289. \begin_layout Standard
  290. Thank you so much for agreeing to read my thesis and give me feedback on
  291. it.
  292. What you are currently reading is a rough draft, in need of many revisions.
  293. You can always find the latest version at
  294. \begin_inset CommandInset href
  295. LatexCommand href
  296. target "https://mneme.dedyn.io/~ryan/Thesis/thesis.pdf"
  297. literal "false"
  298. \end_inset
  299. .
  300. the PDF at this link is updated periodically with my latest revisions,
  301. but you can just download the current version and give me feedback on that.
  302. Don't worry about keeping up with the updates.
  303. \end_layout
  304. \begin_layout Standard
  305. As for what feedback I'm looking for, first of all, don't waste your time
  306. marking spelling mistakes and such.
  307. I haven't run a spell checker on it yet, so let me worry about that.
  308. However, if you see any glaring formatting issues, such as a figure being
  309. too large and getting cut off at the edge of the page, please note them.
  310. In addition, if any of the text in the figures is too small, please note
  311. that as well.
  312. \end_layout
  313. \begin_layout Standard
  314. Beyond that, what I'm mainly interested in is feedback on the content.
  315. For example: does the introduction flow logically, and does it provide
  316. enough background to understand the other chapters? Does each chapter make
  317. it clear what work and analyses I have done? Do the figures clearly communicate
  318. the results I'm trying to show? Do you feel that the claims in the results
  319. and discussion sections are well-supported? Additionally, while I am well
  320. aware that Chapter 1 (the introduction) contains many un-cited claims,
  321. all the other chapters (2,3, and 4)
  322. \emph on
  323. should
  324. \emph default
  325. be fully cited.
  326. So if you notice any un-cited claims in those chapters, please flag them
  327. for my attention.
  328. \end_layout
  329. \begin_layout Standard
  330. You can provide your feedback in whatever way is most convenient to you.
  331. You could mark up this PDF with highlights and notes, then send it back
  332. to me.
  333. Or you could collect your comments in a separate text file and send that
  334. to me, or whatever else you like.
  335. However, if you send me your feedback in a separate document, please note
  336. a section/figure/table number for each comment, and
  337. \emph on
  338. also
  339. \emph default
  340. send me the exact PDF that you read so I can reference it while reading
  341. your comments, since as mentioned above, the current version I'm working
  342. on will have changed by that point (which might include shuffling sections
  343. and figures around, changing their numbers).
  344. One last thing: you'll see a bunch of text in orange boxes throughout the
  345. PDF.
  346. These are notes to myself about things that need to be fixed later, so
  347. if you see a problem noted in an orange box, that means I'm already aware
  348. of it, and there's no need to comment on it.
  349. \end_layout
  350. \begin_layout Standard
  351. My thesis is due Thursday, October 10th, so in order to be useful to me,
  352. I'll need your feedback at least a few days before that, ideally by Monday,
  353. October 7th.
  354. If you have limited time and are unable to get through the whole thesis,
  355. please focus your effors on Chapters 1 and 2, since those are the roughest
  356. and most in need of revision.
  357. Chapter 3 is fairly short and straightforward, and Chapter 4 is an adaptation
  358. of a paper that's already been through a few rounds of revision, so they
  359. should be a lot tighter.
  360. If you can't spare any time between now and then, or if something unexpected
  361. comes up, I understand.
  362. Just let me know.
  363. \end_layout
  364. \begin_layout Standard
  365. Thanks again for your help, and happy reading!
  366. \end_layout
  367. \begin_layout Chapter
  368. Introduction
  369. \end_layout
  370. \begin_layout Section
  371. Background & Significance
  372. \end_layout
  373. \begin_layout Subsection
  374. Biological motivation
  375. \end_layout
  376. \begin_layout Standard
  377. \begin_inset Flex TODO Note (inline)
  378. status open
  379. \begin_layout Plain Layout
  380. Rethink the subsection organization after the intro is written.
  381. \end_layout
  382. \end_inset
  383. \end_layout
  384. \begin_layout Standard
  385. \begin_inset Flex TODO Note (inline)
  386. status open
  387. \begin_layout Plain Layout
  388. Citations are needed all over the place.
  389. A lot of this is knowledge I've just absorbed from years of conversation
  390. in the Salomon lab, without ever having seen a citation for it.
  391. \end_layout
  392. \end_inset
  393. \end_layout
  394. \begin_layout Subsubsection
  395. Rejection is the major long-term threat to organ and tissue allografts
  396. \end_layout
  397. \begin_layout Standard
  398. Organ and tissue transplants are a life-saving treatment for people who
  399. have lost the function of an important organ.
  400. In some cases, it is possible to transplant a patient's own tissue from
  401. one area of their body to another, referred to as an autograft.
  402. This is common for tissues that are distributed throughout many areas of
  403. the body, such as skin and bone.
  404. However, in cases of organ failure, there is no functional self tissue
  405. remaining, and a transplant from another person – the donor – is required.
  406. This is referred to as an allograft.
  407. \end_layout
  408. \begin_layout Standard
  409. \begin_inset Flex TODO Note (inline)
  410. status open
  411. \begin_layout Plain Layout
  412. Possible citation for degree of generic variability: https://www.ncbi.nlm.nih.gov/pu
  413. bmed/22424236?dopt=Abstract
  414. \end_layout
  415. \end_inset
  416. \end_layout
  417. \begin_layout Standard
  418. \begin_inset Flex TODO Note (inline)
  419. status open
  420. \begin_layout Plain Layout
  421. How much mechanistic detail is needed here? My work doesn't really go into
  422. specific rejection mechanisms, so I think it's best to keep it basic.
  423. \end_layout
  424. \end_inset
  425. \end_layout
  426. \begin_layout Standard
  427. Because an allograft comes from a different person, it is genetically distinct
  428. from the rest of the recipient's body.
  429. Some genetic variants occur in protein coding regions, resulting in protein
  430. products that differ from the equivalent proteins in the graft recipient's
  431. own tissue.
  432. As a result, without intervention, the recipient's immune system will eventuall
  433. y identify the graft as foreign tissue and begin attacking it, eventually
  434. resulting in failure and death of the graft, a process referred to as transplan
  435. t rejection.
  436. Rejection is the most significant challenge to the long-term health of
  437. an allograft.
  438. Like any adaptive immune response, graft rejection generally occurs via
  439. two broad mechanisms: cellular immunity, in which CD8+ T-cells induce apoptosis
  440. in the graft cells; and humoral immunity, in which B-cells produce antibodies
  441. that bind to graft proteins and direct an immune response against the graft.
  442. In either case, rejection shows most of the typical hallmarks of an adaptive
  443. immune response, in particular mediation by CD4+ T-cells and formation
  444. of immune memory.
  445. \end_layout
  446. \begin_layout Subsubsection
  447. Diagnosis and treatment of allograft rejection is a major challenge
  448. \end_layout
  449. \begin_layout Standard
  450. To prevent rejection, allograft recipients are treated with immune suppression.
  451. The goal is to achieve sufficient suppression of the immune system to prevent
  452. rejection of the graft without compromising the ability of the immune system
  453. to raise a normal response against infection.
  454. As such, a delicate balance must be struck: insufficient immune suppression
  455. may lead to rejection and ultimately loss of the graft; exceissive suppression
  456. leaves the patient vulnerable to life-threatening opportunistic infections.
  457. Because every patient is different, immune suppression must be tailored
  458. for each patient.
  459. Furthermore, immune suppression must be tuned over time, as the immune
  460. system's activity is not static, nor is it held in a steady state.
  461. In order to properly adjust the dosage of immune suppression drugs, it
  462. is necessary to monitor the health of the transplant and increase the dosage
  463. if evidence of rejection is observed.
  464. \end_layout
  465. \begin_layout Standard
  466. However, diagnosis of rejection is a significant challenge.
  467. Early diagnosis is essential in order to step up immune suppression before
  468. the immune system damages the graft beyond recovery.
  469. The current gold standard test for graft rejection is a tissue biopsy,
  470. examained for visible signs of rejection by a trained histologist.
  471. When a patient shows symptoms of possible rejection, a
  472. \begin_inset Quotes eld
  473. \end_inset
  474. for cause
  475. \begin_inset Quotes erd
  476. \end_inset
  477. biopsy is performed to confirm the diagnosis, and immune suppression is
  478. adjusted as necessary.
  479. However, in many cases, the early stages of rejection are asymptomatic,
  480. known as
  481. \begin_inset Quotes eld
  482. \end_inset
  483. sub-clinical
  484. \begin_inset Quotes erd
  485. \end_inset
  486. rejection.
  487. In light of this, is is now common to perform
  488. \begin_inset Quotes eld
  489. \end_inset
  490. protocol biopsies
  491. \begin_inset Quotes erd
  492. \end_inset
  493. at specific times after transplantation of a graft, even if no symptoms
  494. of rejection are apparent, in addition to
  495. \begin_inset Quotes eld
  496. \end_inset
  497. for cause
  498. \begin_inset Quotes erd
  499. \end_inset
  500. biopsies
  501. \begin_inset CommandInset citation
  502. LatexCommand cite
  503. key "Wilkinson2006"
  504. literal "false"
  505. \end_inset
  506. .
  507. \end_layout
  508. \begin_layout Standard
  509. However, biopsies have a number of downsides that limit their effectiveness
  510. as a diagnostic tool.
  511. First, the need for manual inspection by a histologist means that diagnosis
  512. is subject to the biases of the particular histologist examining the biopsy.
  513. In marginal cases two different histologists may give two different diagnoses
  514. to the same biopsy.
  515. Second, a biopsy can only evaluate if rejection is occurring in the section
  516. of the graft from which the tissue was extracted.
  517. If rejection is only occurring in one section of the graft and the tissue
  518. is extracted from a different section, it may result in a false negative
  519. diagnosis.
  520. Most importantly, however, extraction of tissue from a graft is invasive
  521. and is treated as an injury by the body, which results in inflammation
  522. that in turn promotes increased immune system activity.
  523. Hence, the invasiveness of biopsies severely limits the frequency with
  524. which the can safely be performed.
  525. Typically protocol biopsies are not scheduled more than about once per
  526. month
  527. \begin_inset CommandInset citation
  528. LatexCommand cite
  529. key "Wilkinson2006"
  530. literal "false"
  531. \end_inset
  532. .
  533. A less invasive diagnostic test for rejection would bring manifold benefits.
  534. Such a test would enable more frequent testing and therefore earlier detection
  535. of rejection events.
  536. In addition, having a larger pool of historical data for a given patient
  537. would make it easier to evaluate when a given test is outside the normal
  538. parameters for that specific patient, rather than relying on normal ranges
  539. for the population as a whole.
  540. Lastly, more frequent tests would be a boon to the transplant research
  541. community.
  542. Beyond simply providing more data, the increased time granularity of the
  543. tests will enable studying the progression of a rejection event on the
  544. scale of days to weeks, rather than months.
  545. \end_layout
  546. \begin_layout Subsubsection
  547. Memory cells are resistant to immune suppression
  548. \end_layout
  549. \begin_layout Standard
  550. One of the defining features of the adaptive immune system is immune memory:
  551. the ability of the immune system to recognize a previously encountered
  552. foreign antigen and respond more quickly and more strongly to that antigen
  553. in subsequent encounters.
  554. When the immune system first encounters a new antigen, the lymphocytes
  555. that respond are known as naive cells – T-cells and B-cells that have never
  556. detected their target antigens before.
  557. Once activated by their specific antigen presented by an antigen-presenting
  558. cell in the proper co-stimulatory context, naive cells differentiate into
  559. effector cells that carry out their respective functions in targeting and
  560. destroying the source of the foreign antigen.
  561. The requirement for co-stimulation is an important feature of naive cells
  562. that limits
  563. \begin_inset Quotes eld
  564. \end_inset
  565. false positive
  566. \begin_inset Quotes erd
  567. \end_inset
  568. immune responses, because antigen-presenting cells usually only express
  569. the proper co-stimulation after detecting evidence of an infection, such
  570. as the presence of common bacterial cell components or inflamed tissue.
  571. Most effector cells die after the foreign antigen is cleared, but some
  572. remain and differentiate into memory cells.
  573. Like naive cells, memory cells respond to detection of their specific antigen
  574. by differentiating into effector cells, ready to fight an infection.
  575. However, unlike naive cells, memory cells do not require the same degree
  576. of co-stimulatory signaling for activation, and once activated, they proliferat
  577. e and differentiate into effector cells more quickly than naive cells do.
  578. \end_layout
  579. \begin_layout Standard
  580. In the context of a pathogenic infection, immune memory is a major advantage,
  581. allowing an organism to rapidly fight off a previously encountered pathogen
  582. much more quickly and effectively than the first time it was encountered.
  583. However, if effector cells that recognize an antigen from an allograft
  584. are allowed to differentiate into memory cells, suppressing rejection of
  585. the graft becomes much more difficult.
  586. Many immune suppression drugs work by interfering with the co-stimulation
  587. that naive cells require in order to mount an immune response.
  588. Since memory cells do not require this co-stimulation, these drugs are
  589. not effective at suppressing an immune response that is mediated by memory
  590. cells.
  591. Secondly, because memory cells are able to mount a stronger and faster
  592. response to an antigen, all else being equal they require stronger immune
  593. suppression than naive cells to prevent an immune response.
  594. However, immune suppression affects the entire immune system, not just
  595. cells recognizing a specific antigen, so increasing the dosage of immune
  596. suppression drugs also increases the risk of complications from a compromised
  597. immune system, such as opportunistic infections.
  598. While the differences in cell surface markers between naive and memory
  599. cells have been fairly well characterized, the internal regulatory mechanisms
  600. that allow memory cells to respond more quickly and without co-stimulation
  601. are still poorly understood.
  602. In order to develop immune suppression that either prevents the formation
  603. of memory cells or works more effectively against memory cells, the mechanisms
  604. of immune memory formation and regulation must be better understood.
  605. \end_layout
  606. \begin_layout Subsection
  607. Overview of bioinformatic analysis methods
  608. \end_layout
  609. \begin_layout Standard
  610. The studies presented in this work all involve the analysis of high-throughput
  611. genomic and epigenomic data.
  612. These data present many unique analysis challenges, and a wide array of
  613. software tools are available to analyze them.
  614. This section presents an overview of the methods used, including what problems
  615. they solve, what assumptions they make, and a basic description of how
  616. they work.
  617. \end_layout
  618. \begin_layout Standard
  619. \begin_inset Flex TODO Note (inline)
  620. status open
  621. \begin_layout Plain Layout
  622. Many of these points may also be addressed in the approach/methods sections
  623. of the following chapters? Redundant?
  624. \end_layout
  625. \end_inset
  626. \end_layout
  627. \begin_layout Subsubsection
  628. Limma: The standard linear modeling framework for genomics
  629. \end_layout
  630. \begin_layout Standard
  631. Linear models are a generalization of the
  632. \begin_inset Formula $t$
  633. \end_inset
  634. -test and ANOVA to arbitrarily complex experimental designs.
  635. In a typical linear model, there is one dependent variable observation
  636. per sample.
  637. For example, in a linear model of height as a function of age and sex,
  638. there is one height measurement per person.
  639. However, when analyzing genomic data, each sample consists of observations
  640. of thousands of dependent variables.
  641. For example, in an RNA-seq experiment, the dependent variables may be the
  642. count of RNA-seq reads for each annotated gene.
  643. In abstract terms, each dependent variable being measured is referred to
  644. as a feature.
  645. The simplest approach to analyzing such data would be to fit the same model
  646. independently to each feature.
  647. However, this is undesirable for most genomics data sets.
  648. Genomics assays like high-throughput sequencing are expensive, and often
  649. generating the samples is also quite expensive and time-consuming.
  650. This expense limits the sample sizes typically employed in genomics experiments
  651. , and as a result the statistical power of each individual feature's linear
  652. model is likewise limited.
  653. However, because thousands of features from the same samples are analyzed
  654. together, there is an opportunity to improve the statistical power of the
  655. analysis by exploiting shared patterns of variation across features.
  656. This is the core feature of limma, a linear modeling framework designed
  657. for genomic data.
  658. Limma is typically used to analyze expression microarray data, and more
  659. recently RNA-seq data, but it can also be used to analyze any other data
  660. for which linear modeling is appropriate.
  661. \end_layout
  662. \begin_layout Standard
  663. The central challenge when fitting a linear model is to estimate the variance
  664. of the data accurately.
  665. This quantity is the most difficult to estimate when sample sizes are small.
  666. A single shared variance could be estimated for all of the features together,
  667. and this estimate would be very stable, in contrast to the individual feature
  668. variance estimates.
  669. However, this would require the assumption that every feature is equally
  670. variable, which is known to be false for most genomic data sets.
  671. Limma offers a compromise between these two extremes by using a method
  672. called empirical Bayes moderation to
  673. \begin_inset Quotes eld
  674. \end_inset
  675. squeeze
  676. \begin_inset Quotes erd
  677. \end_inset
  678. the distribution of estimated variances toward a single common value that
  679. represents the variance of an average feature in the data
  680. \begin_inset CommandInset citation
  681. LatexCommand cite
  682. key "Smyth2004"
  683. literal "false"
  684. \end_inset
  685. .
  686. While the individual feature variance estimates are not stable, the common
  687. variance estiamate for the entire data set is quite stable, so using a
  688. combination of the two yields a variance estimate for each feature with
  689. greater precision than the individual feature varaiances.
  690. The trade-off for this improvement is that squeezing each estimated variance
  691. toward the common value introduces some bias – the variance will be underestima
  692. ted for features with high variance and overestimated for features with
  693. low variance.
  694. Essentially, limma assumes that extreme variances are less common than
  695. variances close to the common value.
  696. The variance estimates from this empirical Bayes procedure are shown empiricall
  697. y to yield greater statistical power than either the individual feature
  698. variances or the single common value.
  699. \end_layout
  700. \begin_layout Standard
  701. On top of this core framework, limma also implements many other enhancements
  702. that, further relax the assumptions of the model and extend the scope of
  703. what kinds of data it can analyze.
  704. Instead of squeezing toward a single common variance value, limma can model
  705. the common variance as a function of a covariate, such as average expression
  706. \begin_inset CommandInset citation
  707. LatexCommand cite
  708. key "Law2013"
  709. literal "false"
  710. \end_inset
  711. .
  712. This is essential for RNA-seq data, where higher gene counts yield more
  713. precise expression measurements and therefore smaller variances than low-count
  714. genes.
  715. While linear models typically assume that all samples have equal variance,
  716. limma is able to relax this assumption by identifying and down-weighting
  717. samples the diverge more strongly from the lienar model across many features
  718. \begin_inset CommandInset citation
  719. LatexCommand cite
  720. key "Ritchie2006,Liu2015"
  721. literal "false"
  722. \end_inset
  723. .
  724. In addition, limma is also able to fit simple mixed models incorporating
  725. one random effect in addition to the fixed effects represented by an ordinary
  726. linear model
  727. \begin_inset CommandInset citation
  728. LatexCommand cite
  729. key "Smyth2005a"
  730. literal "false"
  731. \end_inset
  732. .
  733. Once again, limma shares information between features to obtain a robust
  734. estimate for the random effect correlation.
  735. \end_layout
  736. \begin_layout Subsubsection
  737. edgeR provides limma-like analysis features for count data
  738. \end_layout
  739. \begin_layout Standard
  740. Although limma can be applied to read counts from RNA-seq data, it is less
  741. suitable for counts from ChIP-seq data, which tend to be much smaller and
  742. therefore violate the assumption of a normal distribution more severely.
  743. For all count-based data, the edgeR package works similarly to limma, but
  744. uses a generalized linear model instead of a linear model.
  745. The most important difference is that the GLM in edgeR models the counts
  746. directly using a negative binomial distribution rather than modeling the
  747. normalized log counts using a normal distribution
  748. \begin_inset CommandInset citation
  749. LatexCommand cite
  750. key "Chen2014,McCarthy2012,Robinson2010a"
  751. literal "false"
  752. \end_inset
  753. .
  754. The negative binomial is a good fit for count data because it can be derived
  755. as a gamma-distributed mixture of Poisson distributions.
  756. The Poisson distribution accurately represents the distribution of counts
  757. expected for a given gene abundance, and the gamma distribution is then
  758. used to represent the variation in gene abundance between biological replicates.
  759. For this reason, the square root of the dispersion paramter of the negative
  760. binomial is sometimes referred to as the biological coefficient of variation,
  761. since it represents the variability that was present in the samples prior
  762. to the Poisson
  763. \begin_inset Quotes eld
  764. \end_inset
  765. noise
  766. \begin_inset Quotes erd
  767. \end_inset
  768. that was generated by the random sampling of reads in proportion to feature
  769. abundances.
  770. The choice of a gamma distribution is arbitrary and motivated by mathematical
  771. convenience, since a gamma-Poisson mixture yields the numerically tractable
  772. negative binomial distribution.
  773. Thus, edgeR assumes
  774. \emph on
  775. a prioi
  776. \emph default
  777. that the variation in abundances between replicates follows a gamma distribution.
  778. For differential abundance testing, edgeR offers a likelihood ratio test,
  779. but more recently recommends a quasi-likelihood test that properly factors
  780. the uncertainty in variance estimation into the statistical significance
  781. for each feature
  782. \begin_inset CommandInset citation
  783. LatexCommand cite
  784. key "Lund2012"
  785. literal "false"
  786. \end_inset
  787. .
  788. \end_layout
  789. \begin_layout Subsubsection
  790. ChIP-seq Peak calling
  791. \end_layout
  792. \begin_layout Standard
  793. Unlike RNA-seq data, in which gene annotations provide a well-defined set
  794. of genomic regions in which to count reads, ChIP-seq data can potentially
  795. occur anywhere in the genome.
  796. However, most genome regions will not contain significant ChIP-seq read
  797. coverage, and analyzing every position in the entire genome is statistically
  798. and computationally infeasible, so it is necesary to identify regions of
  799. interest inside which ChIP-seq reads will be counted and analyzed.
  800. One option is to define a set of interesting regions
  801. \emph on
  802. a priori
  803. \emph default
  804. , for example by defining a promoter region for each annotated gene.
  805. However, it is also possible to use the ChIP-seq data itself to identify
  806. regions with ChIP-seq read coverage significantly above the background
  807. level, known as peaks.
  808. \end_layout
  809. \begin_layout Standard
  810. There are generally two kinds of peaks that can be identified: narrow peaks
  811. and broadly enriched regions.
  812. Proteins like transcription factors that bind specific sites in the genome
  813. typically show most of their read coverage at these specific sites and
  814. very little coverage anywhere else.
  815. Because the footprint of the protein is consistent wherever it binds, each
  816. peak has a consistent size, typically tens to hundreds of base pairs.
  817. Algorithms like MACS exploit this pattern to identify specific loci at
  818. which such
  819. \begin_inset Quotes eld
  820. \end_inset
  821. narrow peaks
  822. \begin_inset Quotes erd
  823. \end_inset
  824. occur by looking for the characteristic peak shape in the ChIP-seq coverage
  825. rising above the surrounding background coverage
  826. \begin_inset CommandInset citation
  827. LatexCommand cite
  828. key "Zhang2008"
  829. literal "false"
  830. \end_inset
  831. .
  832. In contrast, some proteins, chief among them histones, do not bind only
  833. at a small number of specific sites, but rather bind potentailly almost
  834. everywhere in the entire genome.
  835. When looking at histone marks, adjacent histones tend to be similarly marked,
  836. and a given mark may be present on an arbitrary number of consecutive histones
  837. along the genome.
  838. Hence, there is no consistent
  839. \begin_inset Quotes eld
  840. \end_inset
  841. footprint size
  842. \begin_inset Quotes erd
  843. \end_inset
  844. for ChIP-seq peaks based on histone marks, and peaks typically span many
  845. histones.
  846. Hence, typical peaks span many hundreds or even thousands of base pairs.
  847. Instead of identifying specific loci of strong enrichment, algorithms like
  848. SICER assume that peaks are represented in the ChIP-seq data by modest
  849. enrichment above background occurring across broad regions, and they attempt
  850. to identify the extent of those regions
  851. \begin_inset CommandInset citation
  852. LatexCommand cite
  853. key "Zang2009"
  854. literal "false"
  855. \end_inset
  856. .
  857. In all cases, better results are obtained if the local background coverage
  858. level can be estimated from ChIP-seq input samples, since various biases
  859. can result in uneven background coverage.
  860. \end_layout
  861. \begin_layout Standard
  862. Regardless of the type of peak identified, it is important to identify peaks
  863. that occur consistently across biological replicates.
  864. The ENCODE project has developed a method called irreproducible discovery
  865. rate for this purpose
  866. \begin_inset CommandInset citation
  867. LatexCommand cite
  868. key "Li2006"
  869. literal "false"
  870. \end_inset
  871. .
  872. The IDR is defined as the probability that a peak identified in one biological
  873. replicate will
  874. \emph on
  875. not
  876. \emph default
  877. also be identified in a second replicate.
  878. Where the more familiar false discovery rate measures the degree of corresponde
  879. nce between a data-derived ranked list and the true list of significant
  880. features, IDR instead measures the degree of correspondence between two
  881. ranked lists derived from different data.
  882. IDR assumes that the highest-ranked features are
  883. \begin_inset Quotes eld
  884. \end_inset
  885. signal
  886. \begin_inset Quotes erd
  887. \end_inset
  888. peaks that tend to be listed in the same order in both lists, while the
  889. lowest-ranked features are essentially noise peaks, listed in random order
  890. with no correspondence between the lists.
  891. IDR attempts to locate the
  892. \begin_inset Quotes eld
  893. \end_inset
  894. crossover point
  895. \begin_inset Quotes erd
  896. \end_inset
  897. between the signal and the noise by determining how for down the list the
  898. correspondence between feature ranks breaks down.
  899. \end_layout
  900. \begin_layout Standard
  901. In addition to other considerations, if called peaks are to be used as regions
  902. of interest for differential abundance analysis, then care must be taken
  903. to call peaks in a way that is blind to differential abundance between
  904. experimental conditions, or else the statistical significance calculations
  905. for differential abundance will overstate their confidence in the results.
  906. The csaw package provides guidelines for calling peaks in this way: peaks
  907. are called based on a combination of all ChIP-seq reads from all experimental
  908. conditions, so that the identified peaks are based on the average abundance
  909. across all conditions, which is independent of any differential abundance
  910. between condtions
  911. \begin_inset CommandInset citation
  912. LatexCommand cite
  913. key "Lun2015a"
  914. literal "false"
  915. \end_inset
  916. .
  917. \end_layout
  918. \begin_layout Subsubsection
  919. Normalization of high-throughput data is non-trivial and application-dependant
  920. \end_layout
  921. \begin_layout Standard
  922. High-throughput data sets invariable require some kind of normalization
  923. before further analysis can be conducted.
  924. In general, the goal of normalization is to remove effects in the data
  925. that are caused by technical factors that have nothing to do with the biology
  926. being studied.
  927. \end_layout
  928. \begin_layout Standard
  929. For Affymetrix expression arrays, the standard normalization algorithm used
  930. in most analyses is Robust Multichip Average (RMA).
  931. RMA is designed with the assumption that some fraction of probes on each
  932. array will be artifactual and takes advantage of the fact that each gene
  933. is represented by multiple probes by implementing normalization and summarizati
  934. on steps that are robust against outlier probes.
  935. However, RMA uses the probe intensities of all arrays in the data set in
  936. the normalization of each individual array, meaning that the normalized
  937. expression values in each array depend on every array in the data set,
  938. and will necessarily change each time an array is added or removed from
  939. the data set.
  940. If this is undesirable, frozen RMA implements a variant of RMA where the
  941. relevant distributional parameters are learned from a large reference set
  942. of diverse public array data sets and then
  943. \begin_inset Quotes eld
  944. \end_inset
  945. frozen
  946. \begin_inset Quotes erd
  947. \end_inset
  948. , so that each array is effectively normalized against this frozen reference
  949. set rather than the other arrays in the data set under study.
  950. \end_layout
  951. \begin_layout Standard
  952. In contrast, high-throughput sequencing data present very different normalizatio
  953. n challenges.
  954. The simplest case is RNA-seq in which read counts are obtained for a set
  955. of gene annotations, yielding a matrix of counts with rows representing
  956. genes and columns representing samples.
  957. Because RNA-seq approximates a process of sampling from a population with
  958. replacement, each gene's count is only interpretable as a fraction of the
  959. total reads for that sample.
  960. For that reason, RNA-seq abundances are often reported as counts per million
  961. (CPM).
  962. Furthermore, if the abundance of a single gene increases, then in order
  963. for its fraction of the total reads to increase, all other genes' fractions
  964. must decrease to accomodate it.
  965. This effect is known as composition bias, and it is an artifact of the
  966. read sampling process that has nothing to do with the biology of the samples
  967. and must therefore be normalized out.
  968. The most commonly used methods to normalize for composition bias in RNA-seq
  969. data seek to equalize the average gene abundance across samples, under
  970. the assumption that the average gene is likely not changing
  971. \begin_inset CommandInset citation
  972. LatexCommand cite
  973. key "Robinson2010,Anders2010"
  974. literal "false"
  975. \end_inset
  976. .
  977. \end_layout
  978. \begin_layout Standard
  979. In ChIP-seq data, normalization is not as straightforward.
  980. The csaw package implements several different normalization strategies
  981. and provides guidance on when to use each one
  982. \begin_inset CommandInset citation
  983. LatexCommand cite
  984. key "Lun2015a"
  985. literal "false"
  986. \end_inset
  987. .
  988. Briefly, a typical ChIP-seq sample has a bimodal distribution of read counts:
  989. a low-abundance mode representing background regions and a high-abundance
  990. mode representing signal regions.
  991. This offers two potential normalization targets: equalizing background
  992. coverage or equalizing signal coverage.
  993. If the experiment is well controlled and ChIP efficiency is known to be
  994. consistent across all samples, then normalizing the background coverage
  995. to be equal across all samples is a reasonable strategy.
  996. If this is not a safe assumption, then the preferred strategy is to normalize
  997. the signal regions in a way similar to RNA-seq data by assuming that the
  998. average signal region is not changing abundance between samples.
  999. Beyond this, if a ChIP-seq experiment has a more complicated structure
  1000. that doesn't show the typical bimodal count distribution, it may be necessary
  1001. to implement a normalization as a smooth function of abundance.
  1002. However, this strategy makes a much stronger assumption about the data:
  1003. that the average log fold change is zero across all abundance levels.
  1004. Hence, the simpler scaling normalziations based on background or signal
  1005. regions are generally preferred whenever possible.
  1006. \end_layout
  1007. \begin_layout Subsubsection
  1008. ComBat and SVA for correction of known and unknown batch effects
  1009. \end_layout
  1010. \begin_layout Standard
  1011. In addition to well-understood effects that can be easily normalized out,
  1012. a data set often contains confounding biological effects that must be accounted
  1013. for in the modeling step.
  1014. For instance, in an experiment with pre-treatment and post-treatment samples
  1015. of cells from several different donors, donor variability represents a
  1016. known batch effect.
  1017. The most straightforward correction for known batches is to estimate the
  1018. mean for each batch independently and subtract out the differences, so
  1019. that all batches have identical means for each feature.
  1020. However, as with variance estimation, estimating the differences in batch
  1021. means is not necessarily robust at the feature level, so the ComBat method
  1022. adds empirical Bayes squeezing of the batch mean differences toward a common
  1023. value, analogous to limma's empirical Bayes squeezing of feature variance
  1024. estimates
  1025. \begin_inset CommandInset citation
  1026. LatexCommand cite
  1027. key "Johnson2007"
  1028. literal "false"
  1029. \end_inset
  1030. .
  1031. Effectively, ComBat assumes that modest differences between batch means
  1032. are real batch effects, but extreme differences between batch means are
  1033. more likely to be the result of outlier observations that happen to line
  1034. up with the batches rather than a genuine batch effect.
  1035. The result is a batch correction that is more robust against outliers than
  1036. simple subtraction of mean differences subtraction.
  1037. \end_layout
  1038. \begin_layout Standard
  1039. In some data sets, unknown batch effects may be present due to inherent
  1040. variability in in the data, either caused by technical or biological effects.
  1041. Examples of unknown batch effects include variations in enrichment efficiency
  1042. between ChIP-seq samples, variations in populations of different cell types,
  1043. and the effects of uncontrolled environmental factors on gene expression
  1044. in humans or live animals.
  1045. In an ordinary linear model context, unknown batch effects cannot be inferred
  1046. and must be treated as random noise.
  1047. However, in high-throughput experiments, once again information can be
  1048. shared across features to identify patterns of un-modeled variation that
  1049. are repeated in many features.
  1050. One attractive strategy would be to perform singular value decomposition
  1051. (SVD) on the matrix os linear model residuals (which contain all the un-modeled
  1052. variation in the data) and take the first few singular vectors as batch
  1053. effects.
  1054. While this can be effective, it makes the unreasonable assumption that
  1055. all batch effects are uncorrelated with any of the effects being modeled.
  1056. Surrogate variable analysis (SVA) starts with this approach, but takes
  1057. some additional steps to identify batch effects in the full data that are
  1058. both highly correlated with the singular vectors in the residuals and least
  1059. correlated with the effects of interest
  1060. \begin_inset CommandInset citation
  1061. LatexCommand cite
  1062. key "Leek2007"
  1063. literal "false"
  1064. \end_inset
  1065. .
  1066. Since the final batch effects are estimated from the full data, moderate
  1067. correlations between the batch effects and effects of interest are allowed,
  1068. which gives SVA much more freedom to estimate the true extent of the batch
  1069. effects compared to simple residual SVD.
  1070. Once the surrogate variables are estimated, they can be included as coefficient
  1071. s in the linear model in a similar fashion to known batch effects in order
  1072. to subtract out their effects on each feature's abundance.
  1073. \end_layout
  1074. \begin_layout Subsubsection
  1075. Factor analysis: PCA, MDS, MOFA
  1076. \end_layout
  1077. \begin_layout Standard
  1078. \begin_inset Flex TODO Note (inline)
  1079. status open
  1080. \begin_layout Plain Layout
  1081. Not sure if this merits a subsection here.
  1082. \end_layout
  1083. \end_inset
  1084. \end_layout
  1085. \begin_layout Itemize
  1086. Batch-corrected PCA is informative, but careful application is required
  1087. to avoid bias
  1088. \end_layout
  1089. \begin_layout Section
  1090. Innovation
  1091. \end_layout
  1092. \begin_layout Standard
  1093. \begin_inset Flex TODO Note (inline)
  1094. status open
  1095. \begin_layout Plain Layout
  1096. Is this entire section redundant with the Approach sections of each chapter?
  1097. I'm not really sure what to write here.
  1098. \end_layout
  1099. \end_inset
  1100. \end_layout
  1101. \begin_layout Subsection
  1102. MSC infusion to improve transplant outcomes (prevent/delay rejection)
  1103. \end_layout
  1104. \begin_layout Standard
  1105. \begin_inset Flex TODO Note (inline)
  1106. status open
  1107. \begin_layout Plain Layout
  1108. Do I still talk about this? It's the motivation for chapter 4, but I don't
  1109. actually present any work related to MSCs.
  1110. \end_layout
  1111. \end_inset
  1112. \end_layout
  1113. \begin_layout Itemize
  1114. Demonstrated in mice, but not yet in primates
  1115. \end_layout
  1116. \begin_layout Itemize
  1117. Mechanism currently unknown, but MSC are known to be immune modulatory
  1118. \end_layout
  1119. \begin_layout Itemize
  1120. Characterize MSC response to interferon gamma
  1121. \end_layout
  1122. \begin_layout Itemize
  1123. IFN-g is thought to stimulate their function
  1124. \end_layout
  1125. \begin_layout Itemize
  1126. Test IFN-g treated MSC infusion as a therapy to delay graft rejection in
  1127. cynomolgus monkeys
  1128. \end_layout
  1129. \begin_layout Itemize
  1130. Monitor animals post-transplant using blood RNA-seq at serial time points
  1131. \end_layout
  1132. \begin_layout Subsection
  1133. Investigate dynamics of histone marks in CD4 T-cell activation and memory
  1134. \end_layout
  1135. \begin_layout Itemize
  1136. Previous studies have looked at single snapshots of histone marks
  1137. \end_layout
  1138. \begin_layout Itemize
  1139. Instead, look at changes in histone marks across activation and memory
  1140. \end_layout
  1141. \begin_layout Subsection
  1142. High-throughput sequencing and microarray technologies
  1143. \end_layout
  1144. \begin_layout Itemize
  1145. Powerful methods for assaying gene expression and epigenetics across entire
  1146. genomes
  1147. \end_layout
  1148. \begin_layout Itemize
  1149. Proper analysis requires finding and exploiting systematic genome-wide trends
  1150. \end_layout
  1151. \begin_layout Chapter
  1152. Reproducible genome-wide epigenetic analysis of H3K4 and H3K27 methylation
  1153. in naive and memory CD4 T-cell activation
  1154. \end_layout
  1155. \begin_layout Standard
  1156. \begin_inset Flex TODO Note (inline)
  1157. status open
  1158. \begin_layout Plain Layout
  1159. Chapter author list: Me, Sarah, Dan
  1160. \end_layout
  1161. \end_inset
  1162. \end_layout
  1163. \begin_layout Standard
  1164. \begin_inset Flex TODO Note (inline)
  1165. status open
  1166. \begin_layout Plain Layout
  1167. Need better section titles throughout the entire chapter
  1168. \end_layout
  1169. \end_inset
  1170. \end_layout
  1171. \begin_layout Section
  1172. Approach
  1173. \end_layout
  1174. \begin_layout Standard
  1175. \begin_inset Flex TODO Note (inline)
  1176. status open
  1177. \begin_layout Plain Layout
  1178. Check on the exact correct way to write
  1179. \begin_inset Quotes eld
  1180. \end_inset
  1181. CD4 T-cell
  1182. \begin_inset Quotes erd
  1183. \end_inset
  1184. .
  1185. I think there might be a plus sign somwehere in there now? Also, maybe
  1186. figure out a reasonable way to abbreviate
  1187. \begin_inset Quotes eld
  1188. \end_inset
  1189. naive CD4 T-cells
  1190. \begin_inset Quotes erd
  1191. \end_inset
  1192. and
  1193. \begin_inset Quotes eld
  1194. \end_inset
  1195. memory CD4 T-cells
  1196. \begin_inset Quotes erd
  1197. \end_inset
  1198. .
  1199. \end_layout
  1200. \end_inset
  1201. \end_layout
  1202. \begin_layout Standard
  1203. \begin_inset Flex TODO Note (inline)
  1204. status open
  1205. \begin_layout Plain Layout
  1206. Is it ok to just copy a bunch of citations from the intros to Sarah's papers?
  1207. That feels like cheating somehow.
  1208. \end_layout
  1209. \end_inset
  1210. \end_layout
  1211. \begin_layout Standard
  1212. CD4 T-cells are central to all adaptive immune responses, as well as immune
  1213. memory [CITE?].
  1214. After an infection is cleared, a subset of the naive CD4 T-cells that responded
  1215. to that infection differentiate into memory CD4 T-cells, which are responsible
  1216. for responding to the same pathogen in the future.
  1217. Memory CD4 T-cells are functionally distinct, able to respond to an infection
  1218. more quickly and without the co-stimulation requried by naive CD4 T-cells.
  1219. However, the molecular mechanisms underlying this functional distinction
  1220. are not well-understood.
  1221. Epigenetic regulation via histone modification is thought to play an important
  1222. role, but while many studies have looked at static snapshots of histone
  1223. methylation in T-cells, few studies have looked at the dynamics of histone
  1224. regulation after T-cell activation, nor the differences in histone methylation
  1225. between naive and memory T-cells.
  1226. H3K4me2, H3K4me3 and H3K27me3 are three histone marks thought to be major
  1227. epigenetic regulators of gene expression.
  1228. The goal of the present study is to investigate the role of these histone
  1229. marks in CD4 T-cell activation kinetics and memory differentiation.
  1230. In static snapshots, H3K4me2 and H3K4me3 are often observed in the promoters
  1231. of highly transcribed genes, while H3K27me3 is more often observed in promoters
  1232. of inactive genes with little to no transcription occurring.
  1233. As a result, the two H3K4 marks have been characterized as
  1234. \begin_inset Quotes eld
  1235. \end_inset
  1236. activating
  1237. \begin_inset Quotes erd
  1238. \end_inset
  1239. marks, while H3K27me3 has been characterized as
  1240. \begin_inset Quotes eld
  1241. \end_inset
  1242. deactivating
  1243. \begin_inset Quotes erd
  1244. \end_inset
  1245. .
  1246. Despite these characterizations, the actual causal relationship between
  1247. these histone modifications and gene transcription is complex and likely
  1248. involves positive and negative feedback loops between the two.
  1249. \end_layout
  1250. \begin_layout Standard
  1251. In order to investigate the relationship between gene expression and these
  1252. histone modifications in the context of naive and memory CD4 T-cell activation,
  1253. a previously published data set of combined RNA-seq and ChIP-seq data was
  1254. re-analyzed using up-to-date methods designed to address the specific analysis
  1255. challenges posed by this data set.
  1256. The data set contains naive and memory CD4 T-cell samples in a time course
  1257. before and after activation.
  1258. Like the original analysis, this analysis looks at the dynamics of these
  1259. marks histone marks and compare them to gene expression dynamics at the
  1260. same time points during activation, as well as comapre them between naive
  1261. and memory cells, in hope of discovering evidence of new mechanistic details
  1262. in the interplay between them.
  1263. The original analysis of this data treated each gene promoter as a monolithinc
  1264. unit and mostly assumed that ChIP-seq reads or peaks occuring anywhere
  1265. within a promoter were equivalent, regardless of where they occurred relative
  1266. to the gene structure.
  1267. For an initial analysis of the data, this was a necessary simplifying assumptio
  1268. n.
  1269. The current analysis aims to relax this assumption, first by directly analyzing
  1270. ChIP-seq peaks for differential modification, and second by taking a mor
  1271. granular look at the ChIP-seq read coverage within promoter regions to
  1272. ask whether the location of histone modifications relative to the gene's
  1273. TSS is an important factor, as opposed to simple proximity.
  1274. \end_layout
  1275. \begin_layout Section
  1276. Methods
  1277. \end_layout
  1278. \begin_layout Standard
  1279. \begin_inset Flex TODO Note (inline)
  1280. status open
  1281. \begin_layout Plain Layout
  1282. Look up some more details from the papers (e.g.
  1283. activation method).
  1284. \end_layout
  1285. \end_inset
  1286. \end_layout
  1287. \begin_layout Standard
  1288. A reproducible workflow was written to analyze the raw ChIP-seq and RNA-seq
  1289. data from previous studies
  1290. \begin_inset CommandInset citation
  1291. LatexCommand cite
  1292. key "gh-cd4-csaw,LaMere2016,LaMere2017"
  1293. literal "true"
  1294. \end_inset
  1295. .
  1296. Briefly, this data consists of RNA-seq and ChIP-seq from CD4 T-cells cultured
  1297. from 4 donors.
  1298. From each donor, naive and memory CD4 T-cells were isolated separately.
  1299. Then cultures of both cells were activated [how?], and samples were taken
  1300. at 4 time points: Day 0 (pre-activation), Day 1 (early activation), Day
  1301. 5 (peak activation), and Day 14 (post-activation).
  1302. For each combination of cell type and time point, RNA was isolated and
  1303. sequenced, and ChIP-seq was performed for each of 3 histone marks: H3K4me2,
  1304. H3K4me3, and H3K27me3.
  1305. The ChIP-seq input DNA was also sequenced for each sample.
  1306. The result was 32 samples for each assay.
  1307. \end_layout
  1308. \begin_layout Subsection
  1309. RNA-seq differential expression analysis
  1310. \end_layout
  1311. \begin_layout Standard
  1312. \begin_inset Note Note
  1313. status collapsed
  1314. \begin_layout Plain Layout
  1315. \begin_inset Float figure
  1316. wide false
  1317. sideways false
  1318. status open
  1319. \begin_layout Plain Layout
  1320. \align center
  1321. \begin_inset Float figure
  1322. wide false
  1323. sideways false
  1324. status collapsed
  1325. \begin_layout Plain Layout
  1326. \align center
  1327. \begin_inset Graphics
  1328. filename graphics/CD4-csaw/rnaseq-compare/ensmebl-vs-entrez-star-CROP.png
  1329. lyxscale 25
  1330. width 35col%
  1331. groupId rna-comp-subfig
  1332. \end_inset
  1333. \end_layout
  1334. \begin_layout Plain Layout
  1335. \begin_inset Caption Standard
  1336. \begin_layout Plain Layout
  1337. STAR quantification, Entrez vs Ensembl gene annotation
  1338. \end_layout
  1339. \end_inset
  1340. \end_layout
  1341. \end_inset
  1342. \begin_inset space \qquad{}
  1343. \end_inset
  1344. \begin_inset Float figure
  1345. wide false
  1346. sideways false
  1347. status collapsed
  1348. \begin_layout Plain Layout
  1349. \align center
  1350. \begin_inset Graphics
  1351. filename graphics/CD4-csaw/rnaseq-compare/ensmebl-vs-entrez-shoal-CROP.png
  1352. lyxscale 25
  1353. width 35col%
  1354. groupId rna-comp-subfig
  1355. \end_inset
  1356. \end_layout
  1357. \begin_layout Plain Layout
  1358. \begin_inset Caption Standard
  1359. \begin_layout Plain Layout
  1360. Salmon+Shoal quantification, Entrez vs Ensembl gene annotation
  1361. \end_layout
  1362. \end_inset
  1363. \end_layout
  1364. \end_inset
  1365. \end_layout
  1366. \begin_layout Plain Layout
  1367. \align center
  1368. \begin_inset Float figure
  1369. wide false
  1370. sideways false
  1371. status collapsed
  1372. \begin_layout Plain Layout
  1373. \align center
  1374. \begin_inset Graphics
  1375. filename graphics/CD4-csaw/rnaseq-compare/star-vs-hisat2-CROP.png
  1376. lyxscale 25
  1377. width 35col%
  1378. groupId rna-comp-subfig
  1379. \end_inset
  1380. \end_layout
  1381. \begin_layout Plain Layout
  1382. \begin_inset Caption Standard
  1383. \begin_layout Plain Layout
  1384. STAR vs HISAT2 quantification, Ensembl gene annotation
  1385. \end_layout
  1386. \end_inset
  1387. \end_layout
  1388. \end_inset
  1389. \begin_inset space \qquad{}
  1390. \end_inset
  1391. \begin_inset Float figure
  1392. wide false
  1393. sideways false
  1394. status collapsed
  1395. \begin_layout Plain Layout
  1396. \align center
  1397. \begin_inset Graphics
  1398. filename graphics/CD4-csaw/rnaseq-compare/star-vs-salmon-CROP.png
  1399. lyxscale 25
  1400. width 35col%
  1401. groupId rna-comp-subfig
  1402. \end_inset
  1403. \end_layout
  1404. \begin_layout Plain Layout
  1405. \begin_inset Caption Standard
  1406. \begin_layout Plain Layout
  1407. Salomn vs STAR quantification, Ensembl gene annotation
  1408. \end_layout
  1409. \end_inset
  1410. \end_layout
  1411. \end_inset
  1412. \end_layout
  1413. \begin_layout Plain Layout
  1414. \align center
  1415. \begin_inset Float figure
  1416. wide false
  1417. sideways false
  1418. status collapsed
  1419. \begin_layout Plain Layout
  1420. \align center
  1421. \begin_inset Graphics
  1422. filename graphics/CD4-csaw/rnaseq-compare/salmon-vs-kallisto-CROP.png
  1423. lyxscale 25
  1424. width 35col%
  1425. groupId rna-comp-subfig
  1426. \end_inset
  1427. \end_layout
  1428. \begin_layout Plain Layout
  1429. \begin_inset Caption Standard
  1430. \begin_layout Plain Layout
  1431. Salmon vs Kallisto quantification, Ensembl gene annotation
  1432. \end_layout
  1433. \end_inset
  1434. \end_layout
  1435. \end_inset
  1436. \begin_inset space \qquad{}
  1437. \end_inset
  1438. \begin_inset Float figure
  1439. wide false
  1440. sideways false
  1441. status collapsed
  1442. \begin_layout Plain Layout
  1443. \align center
  1444. \begin_inset Graphics
  1445. filename graphics/CD4-csaw/rnaseq-compare/salmon-vs-shoal-CROP.png
  1446. lyxscale 25
  1447. width 35col%
  1448. groupId rna-comp-subfig
  1449. \end_inset
  1450. \end_layout
  1451. \begin_layout Plain Layout
  1452. \begin_inset Caption Standard
  1453. \begin_layout Plain Layout
  1454. Salmon+Shoal vs Salmon alone, Ensembl gene annotation
  1455. \end_layout
  1456. \end_inset
  1457. \end_layout
  1458. \end_inset
  1459. \end_layout
  1460. \begin_layout Plain Layout
  1461. \begin_inset Caption Standard
  1462. \begin_layout Plain Layout
  1463. \begin_inset CommandInset label
  1464. LatexCommand label
  1465. name "fig:RNA-norm-comp"
  1466. \end_inset
  1467. RNA-seq comparisons
  1468. \end_layout
  1469. \end_inset
  1470. \end_layout
  1471. \end_inset
  1472. \end_layout
  1473. \end_inset
  1474. \end_layout
  1475. \begin_layout Standard
  1476. Sequence reads were retrieved from the Sequence Read Archive (SRA)
  1477. \begin_inset CommandInset citation
  1478. LatexCommand cite
  1479. key "Leinonen2011"
  1480. literal "false"
  1481. \end_inset
  1482. .
  1483. Five different alignment and quantification methods were tested for the
  1484. RNA-seq data
  1485. \begin_inset CommandInset citation
  1486. LatexCommand cite
  1487. key "Dobin2012,Kim2019,Liao2014,Pimentel2016,Patro2017,gh-shoal,gh-hg38-ref"
  1488. literal "false"
  1489. \end_inset
  1490. .
  1491. Each quantification was tested with both Ensembl transcripts and UCSC known
  1492. gene annotations [CITE? Also which versions of each?].
  1493. Comparisons of downstream results from each combination of quantification
  1494. method and reference revealed that all quantifications gave broadly similar
  1495. results for most genes, so shoal with the Ensembl annotation was chosen
  1496. as the method theoretically most likely to partially mitigate some of the
  1497. batch effect in the data.
  1498. \end_layout
  1499. \begin_layout Standard
  1500. \begin_inset Float figure
  1501. wide false
  1502. sideways false
  1503. status collapsed
  1504. \begin_layout Plain Layout
  1505. \align center
  1506. \begin_inset Float figure
  1507. wide false
  1508. sideways false
  1509. status open
  1510. \begin_layout Plain Layout
  1511. \align center
  1512. \begin_inset Graphics
  1513. filename graphics/CD4-csaw/RNA-seq/PCA-no-batchsub-CROP.png
  1514. lyxscale 25
  1515. width 75col%
  1516. groupId rna-pca-subfig
  1517. \end_inset
  1518. \end_layout
  1519. \begin_layout Plain Layout
  1520. \begin_inset Caption Standard
  1521. \begin_layout Plain Layout
  1522. \series bold
  1523. \begin_inset CommandInset label
  1524. LatexCommand label
  1525. name "fig:RNA-PCA-no-batchsub"
  1526. \end_inset
  1527. Before batch correction
  1528. \end_layout
  1529. \end_inset
  1530. \end_layout
  1531. \end_inset
  1532. \end_layout
  1533. \begin_layout Plain Layout
  1534. \align center
  1535. \begin_inset Float figure
  1536. wide false
  1537. sideways false
  1538. status open
  1539. \begin_layout Plain Layout
  1540. \align center
  1541. \begin_inset Graphics
  1542. filename graphics/CD4-csaw/RNA-seq/PCA-combat-batchsub-CROP.png
  1543. lyxscale 25
  1544. width 75col%
  1545. groupId rna-pca-subfig
  1546. \end_inset
  1547. \end_layout
  1548. \begin_layout Plain Layout
  1549. \begin_inset Caption Standard
  1550. \begin_layout Plain Layout
  1551. \series bold
  1552. \begin_inset CommandInset label
  1553. LatexCommand label
  1554. name "fig:RNA-PCA-ComBat-batchsub"
  1555. \end_inset
  1556. After batch correction with ComBat
  1557. \end_layout
  1558. \end_inset
  1559. \end_layout
  1560. \end_inset
  1561. \end_layout
  1562. \begin_layout Plain Layout
  1563. \begin_inset Caption Standard
  1564. \begin_layout Plain Layout
  1565. \series bold
  1566. \begin_inset CommandInset label
  1567. LatexCommand label
  1568. name "fig:RNA-PCA"
  1569. \end_inset
  1570. PCoA plots of RNA-seq data showing effect of batch correction.
  1571. \end_layout
  1572. \end_inset
  1573. \end_layout
  1574. \end_inset
  1575. \end_layout
  1576. \begin_layout Standard
  1577. Due to an error in sample preparation, the RNA from the samples for days
  1578. 0 and 5 were sequenced using a different kit than those for days 1 and
  1579. 14.
  1580. This induced a substantial batch effect in the data due to differences
  1581. in sequencing biases between the two kits, and this batch effect is unfortunate
  1582. ly confounded with the time point variable (Figure
  1583. \begin_inset CommandInset ref
  1584. LatexCommand ref
  1585. reference "fig:RNA-PCA-no-batchsub"
  1586. plural "false"
  1587. caps "false"
  1588. noprefix "false"
  1589. \end_inset
  1590. ).
  1591. To do the best possible analysis with this data, this batch effect was
  1592. subtracted out from the data using ComBat
  1593. \begin_inset CommandInset citation
  1594. LatexCommand cite
  1595. key "Johnson2007"
  1596. literal "false"
  1597. \end_inset
  1598. , ignoring the time point variable due to the confounding with the batch
  1599. variable.
  1600. The result is a marked improvement, but the unavoidable counfounding with
  1601. time point means that certain real patterns of gene expression will be
  1602. indistinguishable from the batch effect and subtracted out as a result.
  1603. Specifically, any
  1604. \begin_inset Quotes eld
  1605. \end_inset
  1606. zig-zag
  1607. \begin_inset Quotes erd
  1608. \end_inset
  1609. pattern, such as a gene whose expression goes up on day 1, down on day
  1610. 5, and back up again on day 14, will be attenuated or eliminated entirely.
  1611. In the context of a T-cell activation time course, it is unlikely that
  1612. many genes of interest will follow such an expression pattern, so this
  1613. loss was deemed an acceptable cost for correcting the batch effect.
  1614. \end_layout
  1615. \begin_layout Standard
  1616. \begin_inset Float figure
  1617. wide false
  1618. sideways false
  1619. status collapsed
  1620. \begin_layout Plain Layout
  1621. \begin_inset Flex TODO Note (inline)
  1622. status open
  1623. \begin_layout Plain Layout
  1624. Just take the top row
  1625. \end_layout
  1626. \end_inset
  1627. \end_layout
  1628. \begin_layout Plain Layout
  1629. \align center
  1630. \begin_inset Graphics
  1631. filename graphics/CD4-csaw/RNA-seq/weights-vs-covars-CROP.png
  1632. lyxscale 25
  1633. width 100col%
  1634. groupId colwidth-raster
  1635. \end_inset
  1636. \end_layout
  1637. \begin_layout Plain Layout
  1638. \begin_inset Caption Standard
  1639. \begin_layout Plain Layout
  1640. \series bold
  1641. \begin_inset CommandInset label
  1642. LatexCommand label
  1643. name "fig:RNA-seq-weights-vs-covars"
  1644. \end_inset
  1645. RNA-seq sample weights, grouped by experimental and technical covariates.
  1646. \end_layout
  1647. \end_inset
  1648. \end_layout
  1649. \end_inset
  1650. \end_layout
  1651. \begin_layout Standard
  1652. However, removing the systematic component of the batch effect still leaves
  1653. the noise component.
  1654. The gene quantifications from the first batch are substantially noisier
  1655. than those in the second batch.
  1656. This analysis corrected for this by using limma's sample weighting method
  1657. to assign lower weights to the noisy samples of batch 1
  1658. \begin_inset CommandInset citation
  1659. LatexCommand cite
  1660. key "Ritchie2006,Liu2015"
  1661. literal "false"
  1662. \end_inset
  1663. .
  1664. The resulting analysis gives an accurate assessment of statistical significance
  1665. for all comparisons, which unfortuantely means a loss of statistical power
  1666. for comparisons involving samples in batch 1.
  1667. \end_layout
  1668. \begin_layout Standard
  1669. In any case, the RNA-seq counts were first normalized using trimmed mean
  1670. of M-values
  1671. \begin_inset CommandInset citation
  1672. LatexCommand cite
  1673. key "Robinson2010"
  1674. literal "false"
  1675. \end_inset
  1676. , converted to normalized logCPM with quality weights using voomWithQualityWeigh
  1677. ts
  1678. \begin_inset CommandInset citation
  1679. LatexCommand cite
  1680. key "Law2013,Liu2015"
  1681. literal "false"
  1682. \end_inset
  1683. , and batch-corrected at this point using ComBat.
  1684. A linear model was fit to the batch-corrected, quality-weighted data for
  1685. each gene using limma, and each gene was tested for differential expression
  1686. using limma's empirical Bayes moderated
  1687. \begin_inset Formula $t$
  1688. \end_inset
  1689. -test
  1690. \begin_inset CommandInset citation
  1691. LatexCommand cite
  1692. key "Smyth2005,Law2013,Phipson2013"
  1693. literal "false"
  1694. \end_inset
  1695. .
  1696. \end_layout
  1697. \begin_layout Subsection
  1698. ChIP-seq differential modification analysis
  1699. \end_layout
  1700. \begin_layout Standard
  1701. \begin_inset Float figure
  1702. wide false
  1703. sideways false
  1704. status collapsed
  1705. \begin_layout Plain Layout
  1706. \align center
  1707. \begin_inset Float figure
  1708. wide false
  1709. sideways false
  1710. status open
  1711. \begin_layout Plain Layout
  1712. \align center
  1713. \begin_inset Graphics
  1714. filename graphics/CD4-csaw/csaw/CCF-plots-noBL-PAGE2-CROP.pdf
  1715. lyxscale 50
  1716. height 40theight%
  1717. groupId ccf-subfig
  1718. \end_inset
  1719. \end_layout
  1720. \begin_layout Plain Layout
  1721. \begin_inset Caption Standard
  1722. \begin_layout Plain Layout
  1723. \series bold
  1724. \begin_inset CommandInset label
  1725. LatexCommand label
  1726. name "fig:CCF-without-blacklist"
  1727. \end_inset
  1728. Cross-correlation plots without removing blacklisted reads.
  1729. \series default
  1730. Without blacklisting, many artifactual peaks are visible in the cross-correlatio
  1731. ns of the ChIP-seq samples, and the peak at the true fragment size (147
  1732. \begin_inset space ~
  1733. \end_inset
  1734. bp) is frequently overshadowed by the artifactual peak at the read length
  1735. (100
  1736. \begin_inset space ~
  1737. \end_inset
  1738. bp).
  1739. \end_layout
  1740. \end_inset
  1741. \end_layout
  1742. \end_inset
  1743. \end_layout
  1744. \begin_layout Plain Layout
  1745. \align center
  1746. \begin_inset Float figure
  1747. wide false
  1748. sideways false
  1749. status open
  1750. \begin_layout Plain Layout
  1751. \align center
  1752. \begin_inset Graphics
  1753. filename graphics/CD4-csaw/csaw/CCF-plots-PAGE2-CROP.pdf
  1754. lyxscale 50
  1755. height 40theight%
  1756. groupId ccf-subfig
  1757. \end_inset
  1758. \end_layout
  1759. \begin_layout Plain Layout
  1760. \begin_inset Caption Standard
  1761. \begin_layout Plain Layout
  1762. \series bold
  1763. \begin_inset CommandInset label
  1764. LatexCommand label
  1765. name "fig:CCF-with-blacklist"
  1766. \end_inset
  1767. Cross-correlation plots with blacklisted reads removed.
  1768. \series default
  1769. After blacklisting, most ChIP-seq samples have clean-looking periodic cross-cor
  1770. relation plots, with the largest peak around 147
  1771. \begin_inset space ~
  1772. \end_inset
  1773. bp, the expected size for a fragment of DNA from a single nucleosome, and
  1774. little to no peak at the read length, 100
  1775. \begin_inset space ~
  1776. \end_inset
  1777. bp.
  1778. \end_layout
  1779. \end_inset
  1780. \end_layout
  1781. \end_inset
  1782. \end_layout
  1783. \begin_layout Plain Layout
  1784. \begin_inset Caption Standard
  1785. \begin_layout Plain Layout
  1786. \series bold
  1787. \begin_inset CommandInset label
  1788. LatexCommand label
  1789. name "fig:CCF-master"
  1790. \end_inset
  1791. Strand cross-correlation plots for ChIP-seq data, before and after blacklisting.
  1792. \end_layout
  1793. \end_inset
  1794. \end_layout
  1795. \end_inset
  1796. \end_layout
  1797. \begin_layout Standard
  1798. \begin_inset Note Note
  1799. status open
  1800. \begin_layout Plain Layout
  1801. \begin_inset Float figure
  1802. wide false
  1803. sideways false
  1804. status collapsed
  1805. \begin_layout Plain Layout
  1806. \align center
  1807. \begin_inset Graphics
  1808. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-sample-MAplot-bins-CROP.png
  1809. lyxscale 25
  1810. width 100col%
  1811. groupId colwidth-raster
  1812. \end_inset
  1813. \end_layout
  1814. \begin_layout Plain Layout
  1815. \begin_inset Caption Standard
  1816. \begin_layout Plain Layout
  1817. \series bold
  1818. \begin_inset CommandInset label
  1819. LatexCommand label
  1820. name "fig:MA-plot-bigbins"
  1821. \end_inset
  1822. MA plot of H3K4me2 read counts in 10kb bins for two arbitrary samples.
  1823. \end_layout
  1824. \end_inset
  1825. \end_layout
  1826. \end_inset
  1827. \end_layout
  1828. \end_inset
  1829. \end_layout
  1830. \begin_layout Standard
  1831. \begin_inset Flex TODO Note (inline)
  1832. status open
  1833. \begin_layout Plain Layout
  1834. Be consistent about use of
  1835. \begin_inset Quotes eld
  1836. \end_inset
  1837. differential binding
  1838. \begin_inset Quotes erd
  1839. \end_inset
  1840. vs
  1841. \begin_inset Quotes eld
  1842. \end_inset
  1843. differential modification
  1844. \begin_inset Quotes erd
  1845. \end_inset
  1846. throughout this chapter.
  1847. The latter is usually preferred.
  1848. \end_layout
  1849. \end_inset
  1850. \end_layout
  1851. \begin_layout Standard
  1852. Sequence reads were retrieved from SRA
  1853. \begin_inset CommandInset citation
  1854. LatexCommand cite
  1855. key "Leinonen2011"
  1856. literal "false"
  1857. \end_inset
  1858. .
  1859. ChIP-seq (and input) reads were aligned to GRCh38 genome assembly using
  1860. Bowtie 2
  1861. \begin_inset CommandInset citation
  1862. LatexCommand cite
  1863. key "Langmead2012,Schneider2017,gh-hg38-ref"
  1864. literal "false"
  1865. \end_inset
  1866. .
  1867. Artifact regions were annotated using a custom implementation of the GreyListCh
  1868. IP algorithm, and these
  1869. \begin_inset Quotes eld
  1870. \end_inset
  1871. greylists
  1872. \begin_inset Quotes erd
  1873. \end_inset
  1874. were merged with the published ENCODE blacklists
  1875. \begin_inset CommandInset citation
  1876. LatexCommand cite
  1877. key "greylistchip,Amemiya2019,Dunham2012,gh-cd4-csaw"
  1878. literal "false"
  1879. \end_inset
  1880. .
  1881. Any read or called peak overlapping one of these regions was regarded as
  1882. artifactual and excluded from downstream analyses.
  1883. Figure
  1884. \begin_inset CommandInset ref
  1885. LatexCommand ref
  1886. reference "fig:CCF-master"
  1887. plural "false"
  1888. caps "false"
  1889. noprefix "false"
  1890. \end_inset
  1891. shows the improvement after blacklisting in the strand cross-correlation
  1892. plots, a common quality control plot for ChIP-seq data.
  1893. Peaks were called using epic, an implementation of the SICER algorithm
  1894. \begin_inset CommandInset citation
  1895. LatexCommand cite
  1896. key "Zang2009,gh-epic"
  1897. literal "false"
  1898. \end_inset
  1899. .
  1900. Peaks were also called separately using MACS, but MACS was determined to
  1901. be a poor fit for the data, and these peak calls are not used in any further
  1902. analyses
  1903. \begin_inset CommandInset citation
  1904. LatexCommand cite
  1905. key "Zhang2008"
  1906. literal "false"
  1907. \end_inset
  1908. .
  1909. Consensus peaks were determined by applying the irreproducible discovery
  1910. rate (IDR) framework
  1911. \begin_inset CommandInset citation
  1912. LatexCommand cite
  1913. key "Li2006,gh-idr"
  1914. literal "false"
  1915. \end_inset
  1916. to find peaks consistently called in the same locations across all 4 donors.
  1917. \end_layout
  1918. \begin_layout Standard
  1919. Promoters were defined by computing the distance from each annotated TSS
  1920. to the nearest called peak and examining the distribution of distances,
  1921. observing that peaks for each histone mark were enriched within a certain
  1922. distance of the TSS.
  1923. For H3K4me2 and H3K4me3, this distance was about 1
  1924. \begin_inset space ~
  1925. \end_inset
  1926. kb, while for H3K27me3 it was 2.5
  1927. \begin_inset space ~
  1928. \end_inset
  1929. kb.
  1930. These distances were used as an
  1931. \begin_inset Quotes eld
  1932. \end_inset
  1933. effective promoter radius
  1934. \begin_inset Quotes erd
  1935. \end_inset
  1936. for each mark.
  1937. The promoter region for each gene was defined as the region of the genome
  1938. within this distance upstream or downstream of the gene's annotated TSS.
  1939. For genes with multiple annotated TSSs, a promoter region was defined for
  1940. each TSS individually, and any promoters that overlapped (due to multiple
  1941. TSSs being closer than 2 times the radius) were merged into one large promoter.
  1942. Thus, some genes had multiple promoters defined, which were each analyzed
  1943. separately for differential modification.
  1944. \end_layout
  1945. \begin_layout Standard
  1946. \begin_inset Float figure
  1947. wide false
  1948. sideways false
  1949. status collapsed
  1950. \begin_layout Plain Layout
  1951. \begin_inset Float figure
  1952. wide false
  1953. sideways false
  1954. status collapsed
  1955. \begin_layout Plain Layout
  1956. \align center
  1957. \begin_inset Graphics
  1958. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-PCA-raw-CROP.png
  1959. lyxscale 25
  1960. width 45col%
  1961. groupId pcoa-subfig
  1962. \end_inset
  1963. \end_layout
  1964. \begin_layout Plain Layout
  1965. \begin_inset Caption Standard
  1966. \begin_layout Plain Layout
  1967. \series bold
  1968. \begin_inset CommandInset label
  1969. LatexCommand label
  1970. name "fig:PCoA-H3K4me2-bad"
  1971. \end_inset
  1972. H3K4me2, no correction
  1973. \end_layout
  1974. \end_inset
  1975. \end_layout
  1976. \end_inset
  1977. \begin_inset space \hfill{}
  1978. \end_inset
  1979. \begin_inset Float figure
  1980. wide false
  1981. sideways false
  1982. status collapsed
  1983. \begin_layout Plain Layout
  1984. \align center
  1985. \begin_inset Graphics
  1986. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-PCA-SVsub-CROP.png
  1987. lyxscale 25
  1988. width 45col%
  1989. groupId pcoa-subfig
  1990. \end_inset
  1991. \end_layout
  1992. \begin_layout Plain Layout
  1993. \begin_inset Caption Standard
  1994. \begin_layout Plain Layout
  1995. \series bold
  1996. \begin_inset CommandInset label
  1997. LatexCommand label
  1998. name "fig:PCoA-H3K4me2-good"
  1999. \end_inset
  2000. H3K4me2, SVs subtracted
  2001. \end_layout
  2002. \end_inset
  2003. \end_layout
  2004. \end_inset
  2005. \end_layout
  2006. \begin_layout Plain Layout
  2007. \begin_inset Float figure
  2008. wide false
  2009. sideways false
  2010. status collapsed
  2011. \begin_layout Plain Layout
  2012. \align center
  2013. \begin_inset Graphics
  2014. filename graphics/CD4-csaw/ChIP-seq/H3K4me3-PCA-raw-CROP.png
  2015. lyxscale 25
  2016. width 45col%
  2017. groupId pcoa-subfig
  2018. \end_inset
  2019. \end_layout
  2020. \begin_layout Plain Layout
  2021. \begin_inset Caption Standard
  2022. \begin_layout Plain Layout
  2023. \series bold
  2024. \begin_inset CommandInset label
  2025. LatexCommand label
  2026. name "fig:PCoA-H3K4me3-bad"
  2027. \end_inset
  2028. H3K4me3, no correction
  2029. \end_layout
  2030. \end_inset
  2031. \end_layout
  2032. \end_inset
  2033. \begin_inset space \hfill{}
  2034. \end_inset
  2035. \begin_inset Float figure
  2036. wide false
  2037. sideways false
  2038. status collapsed
  2039. \begin_layout Plain Layout
  2040. \align center
  2041. \begin_inset Graphics
  2042. filename graphics/CD4-csaw/ChIP-seq/H3K4me3-PCA-SVsub-CROP.png
  2043. lyxscale 25
  2044. width 45col%
  2045. groupId pcoa-subfig
  2046. \end_inset
  2047. \end_layout
  2048. \begin_layout Plain Layout
  2049. \begin_inset Caption Standard
  2050. \begin_layout Plain Layout
  2051. \series bold
  2052. \begin_inset CommandInset label
  2053. LatexCommand label
  2054. name "fig:PCoA-H3K4me3-good"
  2055. \end_inset
  2056. H3K4me3, SVs subtracted
  2057. \end_layout
  2058. \end_inset
  2059. \end_layout
  2060. \end_inset
  2061. \end_layout
  2062. \begin_layout Plain Layout
  2063. \begin_inset Float figure
  2064. wide false
  2065. sideways false
  2066. status collapsed
  2067. \begin_layout Plain Layout
  2068. \align center
  2069. \begin_inset Graphics
  2070. filename graphics/CD4-csaw/ChIP-seq/H3K27me3-PCA-raw-CROP.png
  2071. lyxscale 25
  2072. width 45col%
  2073. groupId pcoa-subfig
  2074. \end_inset
  2075. \end_layout
  2076. \begin_layout Plain Layout
  2077. \begin_inset Caption Standard
  2078. \begin_layout Plain Layout
  2079. \series bold
  2080. \begin_inset CommandInset label
  2081. LatexCommand label
  2082. name "fig:PCoA-H3K27me3-bad"
  2083. \end_inset
  2084. H3K27me3, no correction
  2085. \end_layout
  2086. \end_inset
  2087. \end_layout
  2088. \end_inset
  2089. \begin_inset space \hfill{}
  2090. \end_inset
  2091. \begin_inset Float figure
  2092. wide false
  2093. sideways false
  2094. status collapsed
  2095. \begin_layout Plain Layout
  2096. \align center
  2097. \begin_inset Graphics
  2098. filename graphics/CD4-csaw/ChIP-seq/H3K27me3-PCA-SVsub-CROP.png
  2099. lyxscale 25
  2100. width 45col%
  2101. groupId pcoa-subfig
  2102. \end_inset
  2103. \end_layout
  2104. \begin_layout Plain Layout
  2105. \begin_inset Caption Standard
  2106. \begin_layout Plain Layout
  2107. \series bold
  2108. \begin_inset CommandInset label
  2109. LatexCommand label
  2110. name "fig:PCoA-H3K27me3-good"
  2111. \end_inset
  2112. H3K27me3, SVs subtracted
  2113. \end_layout
  2114. \end_inset
  2115. \end_layout
  2116. \end_inset
  2117. \end_layout
  2118. \begin_layout Plain Layout
  2119. \begin_inset Caption Standard
  2120. \begin_layout Plain Layout
  2121. \series bold
  2122. \begin_inset CommandInset label
  2123. LatexCommand label
  2124. name "fig:PCoA-ChIP"
  2125. \end_inset
  2126. PCoA plots of ChIP-seq sliding window data, before and after subtracting
  2127. surrogate variables (SVs).
  2128. \end_layout
  2129. \end_inset
  2130. \end_layout
  2131. \end_inset
  2132. \end_layout
  2133. \begin_layout Standard
  2134. Reads in promoters, peaks, and sliding windows across the genome were counted
  2135. and normalized using csaw and analyzed for differential modification using
  2136. edgeR
  2137. \begin_inset CommandInset citation
  2138. LatexCommand cite
  2139. key "Lun2014,Lun2015a,Lund2012,Phipson2016"
  2140. literal "false"
  2141. \end_inset
  2142. .
  2143. Unobserved confounding factors in the ChIP-seq data were corrected using
  2144. SVA
  2145. \begin_inset CommandInset citation
  2146. LatexCommand cite
  2147. key "Leek2007,Leek2014"
  2148. literal "false"
  2149. \end_inset
  2150. .
  2151. Principal coordinate plots of the promoter count data for each histone
  2152. mark before and after subtracting surrogate variable effects are shown
  2153. in Figure
  2154. \begin_inset CommandInset ref
  2155. LatexCommand ref
  2156. reference "fig:PCoA-ChIP"
  2157. plural "false"
  2158. caps "false"
  2159. noprefix "false"
  2160. \end_inset
  2161. .
  2162. \end_layout
  2163. \begin_layout Standard
  2164. To investigate whether the location of a peak within the promoter region
  2165. was important,
  2166. \begin_inset Quotes eld
  2167. \end_inset
  2168. relative coverage profiles
  2169. \begin_inset Quotes erd
  2170. \end_inset
  2171. were generated.
  2172. First, 500-bp sliding windows were tiled around each annotated TSS: one
  2173. window centered on the TSS itself, and 10 windows each upstream and downstream,
  2174. thus covering a 10.5-kb region centered on the TSS with 21 windows.
  2175. Reads in each window for each TSS were counted in each sample, and the
  2176. counts were normalized and converted to log CPM as in the differential
  2177. modification analysis.
  2178. Then, the logCPM values within each promoter were normalized to an average
  2179. of zero, such that each window's normalized abundance now represents the
  2180. relative read depth of that window compared to all other windows in the
  2181. same promoter.
  2182. The normalized abundance values for each window in a promoter are collectively
  2183. referred to as that promoter's
  2184. \begin_inset Quotes eld
  2185. \end_inset
  2186. relative coverage profile
  2187. \begin_inset Quotes erd
  2188. \end_inset
  2189. .
  2190. \end_layout
  2191. \begin_layout Subsection
  2192. MOFA recovers biologically relevant variation from blind analysis by correlating
  2193. across datasets
  2194. \end_layout
  2195. \begin_layout Standard
  2196. \begin_inset ERT
  2197. status open
  2198. \begin_layout Plain Layout
  2199. \backslash
  2200. afterpage{
  2201. \end_layout
  2202. \begin_layout Plain Layout
  2203. \backslash
  2204. begin{landscape}
  2205. \end_layout
  2206. \end_inset
  2207. \end_layout
  2208. \begin_layout Standard
  2209. \begin_inset Float figure
  2210. wide false
  2211. sideways false
  2212. status open
  2213. \begin_layout Plain Layout
  2214. \begin_inset Float figure
  2215. wide false
  2216. sideways false
  2217. status open
  2218. \begin_layout Plain Layout
  2219. \align center
  2220. \begin_inset Graphics
  2221. filename graphics/CD4-csaw/MOFA-varExplaiend-matrix-CROP.png
  2222. lyxscale 25
  2223. width 45col%
  2224. groupId mofa-subfig
  2225. \end_inset
  2226. \end_layout
  2227. \begin_layout Plain Layout
  2228. \begin_inset Caption Standard
  2229. \begin_layout Plain Layout
  2230. \series bold
  2231. \begin_inset CommandInset label
  2232. LatexCommand label
  2233. name "fig:mofa-varexplained"
  2234. \end_inset
  2235. Variance explained in each data set by each latent factor estimated by MOFA.
  2236. \series default
  2237. For each latent factor (LF) learned by MOFA, the variance explained by
  2238. that factor in each data set (
  2239. \begin_inset Quotes eld
  2240. \end_inset
  2241. view
  2242. \begin_inset Quotes erd
  2243. \end_inset
  2244. ) is shown by the shading of the cells in the lower section.
  2245. The upper section shows the total fraction of each data set's variance
  2246. that is explained by all LFs combined.
  2247. \end_layout
  2248. \end_inset
  2249. \end_layout
  2250. \end_inset
  2251. \begin_inset space \hfill{}
  2252. \end_inset
  2253. \begin_inset Float figure
  2254. wide false
  2255. sideways false
  2256. status open
  2257. \begin_layout Plain Layout
  2258. \align center
  2259. \begin_inset Graphics
  2260. filename graphics/CD4-csaw/MOFA-LF-scatter-CROP.png
  2261. lyxscale 25
  2262. width 45col%
  2263. groupId mofa-subfig
  2264. \end_inset
  2265. \end_layout
  2266. \begin_layout Plain Layout
  2267. \begin_inset Caption Standard
  2268. \begin_layout Plain Layout
  2269. \series bold
  2270. \begin_inset CommandInset label
  2271. LatexCommand label
  2272. name "fig:mofa-lf-scatter"
  2273. \end_inset
  2274. Scatter plots of specific pairs of MOFA latent factors.
  2275. \series default
  2276. LFs 1, 4, and 5 explain substantial variation in all data sets, so they
  2277. are plotted against each other in order to reveal patterns of variation
  2278. that are shared across all data sets.
  2279. \end_layout
  2280. \end_inset
  2281. \end_layout
  2282. \end_inset
  2283. \end_layout
  2284. \begin_layout Plain Layout
  2285. \begin_inset Caption Standard
  2286. \begin_layout Plain Layout
  2287. \series bold
  2288. \begin_inset CommandInset label
  2289. LatexCommand label
  2290. name "fig:MOFA-master"
  2291. \end_inset
  2292. MOFA latent factors separate technical confounders from
  2293. \end_layout
  2294. \end_inset
  2295. \end_layout
  2296. \end_inset
  2297. \end_layout
  2298. \begin_layout Standard
  2299. \begin_inset ERT
  2300. status open
  2301. \begin_layout Plain Layout
  2302. \backslash
  2303. end{landscape}
  2304. \end_layout
  2305. \begin_layout Plain Layout
  2306. }
  2307. \end_layout
  2308. \end_inset
  2309. \end_layout
  2310. \begin_layout Standard
  2311. MOFA was run on all the ChIP-seq windows overlapping consensus peaks for
  2312. each histone mark, as well as the RNA-seq data, in order to identify patterns
  2313. of coordinated variation across all data sets
  2314. \begin_inset CommandInset citation
  2315. LatexCommand cite
  2316. key "Argelaguet2018"
  2317. literal "false"
  2318. \end_inset
  2319. .
  2320. The results are summarized in Figure
  2321. \begin_inset CommandInset ref
  2322. LatexCommand ref
  2323. reference "fig:MOFA-master"
  2324. plural "false"
  2325. caps "false"
  2326. noprefix "false"
  2327. \end_inset
  2328. .
  2329. Latent factors 1, 4, and 5 were determined to explain the most variation
  2330. consistently across all data sets (Fgure
  2331. \begin_inset CommandInset ref
  2332. LatexCommand ref
  2333. reference "fig:mofa-varexplained"
  2334. plural "false"
  2335. caps "false"
  2336. noprefix "false"
  2337. \end_inset
  2338. ), and scatter plots of these factors show that they also correlate best
  2339. with the experimental factors (Figure
  2340. \begin_inset CommandInset ref
  2341. LatexCommand ref
  2342. reference "fig:mofa-lf-scatter"
  2343. plural "false"
  2344. caps "false"
  2345. noprefix "false"
  2346. \end_inset
  2347. ).
  2348. Latent factor 2 captures the batch effect in the RNA-seq data.
  2349. Removing the effect of LF2 using MOFA theoretically yields a batch correction
  2350. that does not depend on knowing the experimental factors.
  2351. When this was attempted, the resulting batch correction was comparable
  2352. to ComBat (see Figure
  2353. \begin_inset CommandInset ref
  2354. LatexCommand ref
  2355. reference "fig:RNA-PCA-ComBat-batchsub"
  2356. plural "false"
  2357. caps "false"
  2358. noprefix "false"
  2359. \end_inset
  2360. ), indicating that the ComBat-based batch correction has little room for
  2361. improvement given the problems with the data set.
  2362. \end_layout
  2363. \begin_layout Standard
  2364. \begin_inset Note Note
  2365. status collapsed
  2366. \begin_layout Plain Layout
  2367. \begin_inset Float figure
  2368. wide false
  2369. sideways false
  2370. status open
  2371. \begin_layout Plain Layout
  2372. \align center
  2373. \begin_inset Graphics
  2374. filename graphics/CD4-csaw/MOFA-batch-correct-CROP.png
  2375. lyxscale 25
  2376. width 100col%
  2377. groupId colwidth-raster
  2378. \end_inset
  2379. \end_layout
  2380. \begin_layout Plain Layout
  2381. \begin_inset Caption Standard
  2382. \begin_layout Plain Layout
  2383. \series bold
  2384. \begin_inset CommandInset label
  2385. LatexCommand label
  2386. name "fig:mofa-batchsub"
  2387. \end_inset
  2388. Result of RNA-seq batch-correction using MOFA latent factors
  2389. \end_layout
  2390. \end_inset
  2391. \end_layout
  2392. \end_inset
  2393. \end_layout
  2394. \end_inset
  2395. \end_layout
  2396. \begin_layout Section
  2397. Results
  2398. \end_layout
  2399. \begin_layout Standard
  2400. \begin_inset Flex TODO Note (inline)
  2401. status open
  2402. \begin_layout Plain Layout
  2403. Focus on what hypotheses were tested, then select figures that show how
  2404. those hypotheses were tested, even if the result is a negative.
  2405. Not every interesting result needs to be in here.
  2406. Chapter should tell a story.
  2407. \end_layout
  2408. \end_inset
  2409. \end_layout
  2410. \begin_layout Standard
  2411. \begin_inset Flex TODO Note (inline)
  2412. status open
  2413. \begin_layout Plain Layout
  2414. Maybe reorder these sections to do RNA-seq, then ChIP-seq, then combined
  2415. analyses?
  2416. \end_layout
  2417. \end_inset
  2418. \end_layout
  2419. \begin_layout Subsection
  2420. Interpretation of RNA-seq analysis is limited by a major confounding factor
  2421. \end_layout
  2422. \begin_layout Standard
  2423. \begin_inset Float table
  2424. wide false
  2425. sideways false
  2426. status collapsed
  2427. \begin_layout Plain Layout
  2428. \align center
  2429. \begin_inset Tabular
  2430. <lyxtabular version="3" rows="11" columns="3">
  2431. <features tabularvalignment="middle">
  2432. <column alignment="center" valignment="top">
  2433. <column alignment="center" valignment="top">
  2434. <column alignment="center" valignment="top">
  2435. <row>
  2436. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2437. \begin_inset Text
  2438. \begin_layout Plain Layout
  2439. Test
  2440. \end_layout
  2441. \end_inset
  2442. </cell>
  2443. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2444. \begin_inset Text
  2445. \begin_layout Plain Layout
  2446. Est.
  2447. non-null
  2448. \end_layout
  2449. \end_inset
  2450. </cell>
  2451. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  2452. \begin_inset Text
  2453. \begin_layout Plain Layout
  2454. \begin_inset Formula $\mathrm{FDR}\le10\%$
  2455. \end_inset
  2456. \end_layout
  2457. \end_inset
  2458. </cell>
  2459. </row>
  2460. <row>
  2461. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2462. \begin_inset Text
  2463. \begin_layout Plain Layout
  2464. Naive Day 0 vs Day 1
  2465. \end_layout
  2466. \end_inset
  2467. </cell>
  2468. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2469. \begin_inset Text
  2470. \begin_layout Plain Layout
  2471. 5992
  2472. \end_layout
  2473. \end_inset
  2474. </cell>
  2475. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2476. \begin_inset Text
  2477. \begin_layout Plain Layout
  2478. 1613
  2479. \end_layout
  2480. \end_inset
  2481. </cell>
  2482. </row>
  2483. <row>
  2484. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2485. \begin_inset Text
  2486. \begin_layout Plain Layout
  2487. Naive Day 0 vs Day 5
  2488. \end_layout
  2489. \end_inset
  2490. </cell>
  2491. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2492. \begin_inset Text
  2493. \begin_layout Plain Layout
  2494. 3038
  2495. \end_layout
  2496. \end_inset
  2497. </cell>
  2498. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2499. \begin_inset Text
  2500. \begin_layout Plain Layout
  2501. 32
  2502. \end_layout
  2503. \end_inset
  2504. </cell>
  2505. </row>
  2506. <row>
  2507. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2508. \begin_inset Text
  2509. \begin_layout Plain Layout
  2510. Naive Day 0 vs Day 14
  2511. \end_layout
  2512. \end_inset
  2513. </cell>
  2514. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2515. \begin_inset Text
  2516. \begin_layout Plain Layout
  2517. 1870
  2518. \end_layout
  2519. \end_inset
  2520. </cell>
  2521. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2522. \begin_inset Text
  2523. \begin_layout Plain Layout
  2524. 190
  2525. \end_layout
  2526. \end_inset
  2527. </cell>
  2528. </row>
  2529. <row>
  2530. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2531. \begin_inset Text
  2532. \begin_layout Plain Layout
  2533. Memory Day 0 vs Day 1
  2534. \end_layout
  2535. \end_inset
  2536. </cell>
  2537. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2538. \begin_inset Text
  2539. \begin_layout Plain Layout
  2540. 3195
  2541. \end_layout
  2542. \end_inset
  2543. </cell>
  2544. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2545. \begin_inset Text
  2546. \begin_layout Plain Layout
  2547. 411
  2548. \end_layout
  2549. \end_inset
  2550. </cell>
  2551. </row>
  2552. <row>
  2553. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2554. \begin_inset Text
  2555. \begin_layout Plain Layout
  2556. Memory Day 0 vs Day 5
  2557. \end_layout
  2558. \end_inset
  2559. </cell>
  2560. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2561. \begin_inset Text
  2562. \begin_layout Plain Layout
  2563. 2688
  2564. \end_layout
  2565. \end_inset
  2566. </cell>
  2567. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2568. \begin_inset Text
  2569. \begin_layout Plain Layout
  2570. 18
  2571. \end_layout
  2572. \end_inset
  2573. </cell>
  2574. </row>
  2575. <row>
  2576. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2577. \begin_inset Text
  2578. \begin_layout Plain Layout
  2579. Memory Day 0 vs Day 14
  2580. \end_layout
  2581. \end_inset
  2582. </cell>
  2583. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2584. \begin_inset Text
  2585. \begin_layout Plain Layout
  2586. 1911
  2587. \end_layout
  2588. \end_inset
  2589. </cell>
  2590. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2591. \begin_inset Text
  2592. \begin_layout Plain Layout
  2593. 227
  2594. \end_layout
  2595. \end_inset
  2596. </cell>
  2597. </row>
  2598. <row>
  2599. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2600. \begin_inset Text
  2601. \begin_layout Plain Layout
  2602. Day 0 Naive vs Memory
  2603. \end_layout
  2604. \end_inset
  2605. </cell>
  2606. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2607. \begin_inset Text
  2608. \begin_layout Plain Layout
  2609. 0
  2610. \end_layout
  2611. \end_inset
  2612. </cell>
  2613. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2614. \begin_inset Text
  2615. \begin_layout Plain Layout
  2616. 2
  2617. \end_layout
  2618. \end_inset
  2619. </cell>
  2620. </row>
  2621. <row>
  2622. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2623. \begin_inset Text
  2624. \begin_layout Plain Layout
  2625. Day 1 Naive vs Memory
  2626. \end_layout
  2627. \end_inset
  2628. </cell>
  2629. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2630. \begin_inset Text
  2631. \begin_layout Plain Layout
  2632. 9167
  2633. \end_layout
  2634. \end_inset
  2635. </cell>
  2636. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2637. \begin_inset Text
  2638. \begin_layout Plain Layout
  2639. 5532
  2640. \end_layout
  2641. \end_inset
  2642. </cell>
  2643. </row>
  2644. <row>
  2645. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2646. \begin_inset Text
  2647. \begin_layout Plain Layout
  2648. Day 5 Naive vs Memory
  2649. \end_layout
  2650. \end_inset
  2651. </cell>
  2652. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2653. \begin_inset Text
  2654. \begin_layout Plain Layout
  2655. 0
  2656. \end_layout
  2657. \end_inset
  2658. </cell>
  2659. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2660. \begin_inset Text
  2661. \begin_layout Plain Layout
  2662. 0
  2663. \end_layout
  2664. \end_inset
  2665. </cell>
  2666. </row>
  2667. <row>
  2668. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2669. \begin_inset Text
  2670. \begin_layout Plain Layout
  2671. Day 14 Naive vs Memory
  2672. \end_layout
  2673. \end_inset
  2674. </cell>
  2675. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2676. \begin_inset Text
  2677. \begin_layout Plain Layout
  2678. 6446
  2679. \end_layout
  2680. \end_inset
  2681. </cell>
  2682. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  2683. \begin_inset Text
  2684. \begin_layout Plain Layout
  2685. 2319
  2686. \end_layout
  2687. \end_inset
  2688. </cell>
  2689. </row>
  2690. </lyxtabular>
  2691. \end_inset
  2692. \end_layout
  2693. \begin_layout Plain Layout
  2694. \begin_inset Caption Standard
  2695. \begin_layout Plain Layout
  2696. \series bold
  2697. \begin_inset CommandInset label
  2698. LatexCommand label
  2699. name "tab:Estimated-and-detected-rnaseq"
  2700. \end_inset
  2701. Estimated and detected differentially expressed genes.
  2702. \series default
  2703. \begin_inset Quotes eld
  2704. \end_inset
  2705. Test
  2706. \begin_inset Quotes erd
  2707. \end_inset
  2708. : Which sample groups were compared;
  2709. \begin_inset Quotes eld
  2710. \end_inset
  2711. Est non-null
  2712. \begin_inset Quotes erd
  2713. \end_inset
  2714. : Estimated number of differentially expressed genes, using the method of
  2715. averaging local FDR values
  2716. \begin_inset CommandInset citation
  2717. LatexCommand cite
  2718. key "Phipson2013Thesis"
  2719. literal "false"
  2720. \end_inset
  2721. ;
  2722. \begin_inset Quotes eld
  2723. \end_inset
  2724. \begin_inset Formula $\mathrm{FDR}\le10\%$
  2725. \end_inset
  2726. \begin_inset Quotes erd
  2727. \end_inset
  2728. : Number of significantly differentially expressed genes at an FDR threshold
  2729. of 10%.
  2730. The total number of genes tested was 16707.
  2731. \end_layout
  2732. \end_inset
  2733. \end_layout
  2734. \end_inset
  2735. \end_layout
  2736. \begin_layout Standard
  2737. \begin_inset Float figure
  2738. wide false
  2739. sideways false
  2740. status collapsed
  2741. \begin_layout Plain Layout
  2742. \align center
  2743. \begin_inset Graphics
  2744. filename graphics/CD4-csaw/RNA-seq/PCA-final-12-CROP.png
  2745. lyxscale 25
  2746. width 100col%
  2747. groupId colwidth-raster
  2748. \end_inset
  2749. \end_layout
  2750. \begin_layout Plain Layout
  2751. \begin_inset Caption Standard
  2752. \begin_layout Plain Layout
  2753. \series bold
  2754. \begin_inset CommandInset label
  2755. LatexCommand label
  2756. name "fig:rna-pca-final"
  2757. \end_inset
  2758. PCoA plot of RNA-seq samples after ComBat batch correction.
  2759. \series default
  2760. Each point represents an individual sample.
  2761. Samples with the same combination of cell type and time point are encircled
  2762. with a shaded region to aid in visual identification of the sample groups.
  2763. Samples with of same cell type from the same donor are connected by lines
  2764. to indicate the
  2765. \begin_inset Quotes eld
  2766. \end_inset
  2767. trajectory
  2768. \begin_inset Quotes erd
  2769. \end_inset
  2770. of each donor's cells over time in PCoA space.
  2771. \end_layout
  2772. \end_inset
  2773. \end_layout
  2774. \begin_layout Plain Layout
  2775. \end_layout
  2776. \end_inset
  2777. \end_layout
  2778. \begin_layout Standard
  2779. Genes called present in the RNA-seq data were tested for differential expression
  2780. between all time points and cell types.
  2781. The counts of differentially expressed genes are shown in Table
  2782. \begin_inset CommandInset ref
  2783. LatexCommand ref
  2784. reference "tab:Estimated-and-detected-rnaseq"
  2785. plural "false"
  2786. caps "false"
  2787. noprefix "false"
  2788. \end_inset
  2789. .
  2790. Notably, all the results for Day 0 and Day 5 have substantially fewer genes
  2791. called differentially expressed than any of the results for other time
  2792. points.
  2793. This is an unfortunate result of the difference in sample quality between
  2794. the two batches of RNA-seq data.
  2795. All the samples in Batch 1, which includes all the samples from Days 0
  2796. and 5, have substantially more variability than the samples in Batch 2,
  2797. which includes the other time points.
  2798. This is reflected in the substantially higher weights assigned to Batch
  2799. 2 (Figure
  2800. \begin_inset CommandInset ref
  2801. LatexCommand ref
  2802. reference "fig:RNA-seq-weights-vs-covars"
  2803. plural "false"
  2804. caps "false"
  2805. noprefix "false"
  2806. \end_inset
  2807. ).
  2808. The batch effect has both a systematic component and a random noise component.
  2809. While the systematic component was subtracted out using ComBat (Figure
  2810. \begin_inset CommandInset ref
  2811. LatexCommand ref
  2812. reference "fig:RNA-PCA"
  2813. plural "false"
  2814. caps "false"
  2815. noprefix "false"
  2816. \end_inset
  2817. ), no such correction is possible for the noise component: Batch 1 simply
  2818. has substantially more random noise in it, which reduces the statistical
  2819. power for any differential expression tests involving samples in that batch.
  2820. \end_layout
  2821. \begin_layout Standard
  2822. Despite the difficulty in detecting specific differentially expressed genes,
  2823. there is still evidence that differential expression is present for these
  2824. time points.
  2825. In Figure
  2826. \begin_inset CommandInset ref
  2827. LatexCommand ref
  2828. reference "fig:rna-pca-final"
  2829. plural "false"
  2830. caps "false"
  2831. noprefix "false"
  2832. \end_inset
  2833. , there is a clear separation between naive and memory samples at Day 0,
  2834. despite the fact that only 2 genes were significantly differentially expressed
  2835. for this comparison.
  2836. Similarly, the small numbers of genes detected for the Day 0 vs Day 5 compariso
  2837. ns do not reflect the large separation between these time points in Figure
  2838. \begin_inset CommandInset ref
  2839. LatexCommand ref
  2840. reference "fig:rna-pca-final"
  2841. plural "false"
  2842. caps "false"
  2843. noprefix "false"
  2844. \end_inset
  2845. .
  2846. In addition, the MOFA latent factor plots in Figure
  2847. \begin_inset CommandInset ref
  2848. LatexCommand ref
  2849. reference "fig:mofa-lf-scatter"
  2850. plural "false"
  2851. caps "false"
  2852. noprefix "false"
  2853. \end_inset
  2854. .
  2855. This suggests that there is indeed a differential expression signal present
  2856. in the data for these comparisons, but the large variability in the Batch
  2857. 1 samples obfuscates this signal at the individual gene level.
  2858. As a result, it is impossible to make any meaningful statements about the
  2859. \begin_inset Quotes eld
  2860. \end_inset
  2861. size
  2862. \begin_inset Quotes erd
  2863. \end_inset
  2864. of the gene signature for any time point, since the number of significant
  2865. genes as well as the estimated number of differentially expressed genes
  2866. depends so strongly on the variations in sample quality in addition to
  2867. the size of the differential expression signal in the data.
  2868. Gene-set enrichment analyses are similarly impractical.
  2869. However, analyses looking at genome-wide patterns of expression are still
  2870. practical.
  2871. \end_layout
  2872. \begin_layout Subsection
  2873. H3K4 and H3K27 methylation occur in broad regions and are enriched near
  2874. promoters
  2875. \end_layout
  2876. \begin_layout Standard
  2877. \begin_inset Float table
  2878. wide false
  2879. sideways false
  2880. status collapsed
  2881. \begin_layout Plain Layout
  2882. \align center
  2883. \begin_inset Flex TODO Note (inline)
  2884. status open
  2885. \begin_layout Plain Layout
  2886. Also get
  2887. \emph on
  2888. median
  2889. \emph default
  2890. peak width and maybe other quantiles (25%, 75%)
  2891. \end_layout
  2892. \end_inset
  2893. \end_layout
  2894. \begin_layout Plain Layout
  2895. \align center
  2896. \begin_inset Tabular
  2897. <lyxtabular version="3" rows="4" columns="5">
  2898. <features tabularvalignment="middle">
  2899. <column alignment="center" valignment="top">
  2900. <column alignment="center" valignment="top">
  2901. <column alignment="center" valignment="top">
  2902. <column alignment="center" valignment="top">
  2903. <column alignment="center" valignment="top">
  2904. <row>
  2905. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2906. \begin_inset Text
  2907. \begin_layout Plain Layout
  2908. Histone Mark
  2909. \end_layout
  2910. \end_inset
  2911. </cell>
  2912. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2913. \begin_inset Text
  2914. \begin_layout Plain Layout
  2915. # Peaks
  2916. \end_layout
  2917. \end_inset
  2918. </cell>
  2919. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2920. \begin_inset Text
  2921. \begin_layout Plain Layout
  2922. Mean peak width
  2923. \end_layout
  2924. \end_inset
  2925. </cell>
  2926. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2927. \begin_inset Text
  2928. \begin_layout Plain Layout
  2929. genome coverage
  2930. \end_layout
  2931. \end_inset
  2932. </cell>
  2933. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  2934. \begin_inset Text
  2935. \begin_layout Plain Layout
  2936. FRiP
  2937. \end_layout
  2938. \end_inset
  2939. </cell>
  2940. </row>
  2941. <row>
  2942. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2943. \begin_inset Text
  2944. \begin_layout Plain Layout
  2945. H3K4me2
  2946. \end_layout
  2947. \end_inset
  2948. </cell>
  2949. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2950. \begin_inset Text
  2951. \begin_layout Plain Layout
  2952. 14965
  2953. \end_layout
  2954. \end_inset
  2955. </cell>
  2956. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2957. \begin_inset Text
  2958. \begin_layout Plain Layout
  2959. 3970
  2960. \end_layout
  2961. \end_inset
  2962. </cell>
  2963. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2964. \begin_inset Text
  2965. \begin_layout Plain Layout
  2966. 1.92%
  2967. \end_layout
  2968. \end_inset
  2969. </cell>
  2970. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2971. \begin_inset Text
  2972. \begin_layout Plain Layout
  2973. 14.2%
  2974. \end_layout
  2975. \end_inset
  2976. </cell>
  2977. </row>
  2978. <row>
  2979. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2980. \begin_inset Text
  2981. \begin_layout Plain Layout
  2982. H3K4me3
  2983. \end_layout
  2984. \end_inset
  2985. </cell>
  2986. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2987. \begin_inset Text
  2988. \begin_layout Plain Layout
  2989. 6163
  2990. \end_layout
  2991. \end_inset
  2992. </cell>
  2993. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2994. \begin_inset Text
  2995. \begin_layout Plain Layout
  2996. 2946
  2997. \end_layout
  2998. \end_inset
  2999. </cell>
  3000. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3001. \begin_inset Text
  3002. \begin_layout Plain Layout
  3003. 0.588%
  3004. \end_layout
  3005. \end_inset
  3006. </cell>
  3007. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3008. \begin_inset Text
  3009. \begin_layout Plain Layout
  3010. 6.57%
  3011. \end_layout
  3012. \end_inset
  3013. </cell>
  3014. </row>
  3015. <row>
  3016. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3017. \begin_inset Text
  3018. \begin_layout Plain Layout
  3019. H3K27me3
  3020. \end_layout
  3021. \end_inset
  3022. </cell>
  3023. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3024. \begin_inset Text
  3025. \begin_layout Plain Layout
  3026. 18139
  3027. \end_layout
  3028. \end_inset
  3029. </cell>
  3030. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3031. \begin_inset Text
  3032. \begin_layout Plain Layout
  3033. 18967
  3034. \end_layout
  3035. \end_inset
  3036. </cell>
  3037. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3038. \begin_inset Text
  3039. \begin_layout Plain Layout
  3040. 11.1%
  3041. \end_layout
  3042. \end_inset
  3043. </cell>
  3044. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3045. \begin_inset Text
  3046. \begin_layout Plain Layout
  3047. 22.5%
  3048. \end_layout
  3049. \end_inset
  3050. </cell>
  3051. </row>
  3052. </lyxtabular>
  3053. \end_inset
  3054. \end_layout
  3055. \begin_layout Plain Layout
  3056. \begin_inset Caption Standard
  3057. \begin_layout Plain Layout
  3058. \series bold
  3059. \begin_inset CommandInset label
  3060. LatexCommand label
  3061. name "tab:peak-calling-summary"
  3062. \end_inset
  3063. Peak-calling summary.
  3064. \series default
  3065. For each histone mark, the number of peaks called using SICER at an IDR
  3066. threshold of ???, the mean width of those peaks, the fraction of the genome
  3067. covered by peaks, and the fraction of reads in peaks (FRiP).
  3068. \end_layout
  3069. \end_inset
  3070. \end_layout
  3071. \end_inset
  3072. \end_layout
  3073. \begin_layout Standard
  3074. Table
  3075. \begin_inset CommandInset ref
  3076. LatexCommand ref
  3077. reference "tab:peak-calling-summary"
  3078. plural "false"
  3079. caps "false"
  3080. noprefix "false"
  3081. \end_inset
  3082. gives a summary of the peak calling statistics for each histone mark.
  3083. Consistent with previous observations [CITATION NEEDED], all 3 histone
  3084. marks occur in broad regions spanning many consecutive nucleosomes, rather
  3085. than in sharp peaks as would be expected for a transcription factor or
  3086. other molecule that binds to specific sites.
  3087. This conclusion is further supported by Figure
  3088. \begin_inset CommandInset ref
  3089. LatexCommand ref
  3090. reference "fig:CCF-with-blacklist"
  3091. plural "false"
  3092. caps "false"
  3093. noprefix "false"
  3094. \end_inset
  3095. , in which a clear nucleosome-sized periodicity is visible in the cross-correlat
  3096. ion value for each sample, indicating that each time a given mark is present
  3097. on one histone, it is also likely to be found on adjacent histones as well.
  3098. H3K27me3 enrichment in particular is substantially more broad than either
  3099. H3K4 mark, with a mean peak width of almost 19,000 bp.
  3100. This is also reflected in the periodicity observed in Figure
  3101. \begin_inset CommandInset ref
  3102. LatexCommand ref
  3103. reference "fig:CCF-with-blacklist"
  3104. plural "false"
  3105. caps "false"
  3106. noprefix "false"
  3107. \end_inset
  3108. , which remains strong much farther out for H3K27me3 than the other marks,
  3109. showing H3K27me3 especially tends to be found on long runs of consecutive
  3110. histones.
  3111. \end_layout
  3112. \begin_layout Standard
  3113. \begin_inset Float figure
  3114. wide false
  3115. sideways false
  3116. status open
  3117. \begin_layout Plain Layout
  3118. \begin_inset Flex TODO Note (inline)
  3119. status open
  3120. \begin_layout Plain Layout
  3121. Ensure this figure uses the peak calls from the new analysis.
  3122. \end_layout
  3123. \end_inset
  3124. \end_layout
  3125. \begin_layout Plain Layout
  3126. \begin_inset Flex TODO Note (inline)
  3127. status open
  3128. \begin_layout Plain Layout
  3129. Need a control: shuffle all peaks and repeat, N times.
  3130. Do real vs shuffled control both in a top/bottom arrangement.
  3131. \end_layout
  3132. \end_inset
  3133. \end_layout
  3134. \begin_layout Plain Layout
  3135. \begin_inset Flex TODO Note (inline)
  3136. status open
  3137. \begin_layout Plain Layout
  3138. Consider counting TSS inside peaks as negative number indicating how far
  3139. \emph on
  3140. inside
  3141. \emph default
  3142. the peak the TSS is (i.e.
  3143. distance to nearest non-peak area).
  3144. \end_layout
  3145. \end_inset
  3146. \end_layout
  3147. \begin_layout Plain Layout
  3148. \begin_inset Flex TODO Note (inline)
  3149. status open
  3150. \begin_layout Plain Layout
  3151. The H3K4 part of this figure is included in
  3152. \begin_inset CommandInset citation
  3153. LatexCommand cite
  3154. key "LaMere2016"
  3155. literal "false"
  3156. \end_inset
  3157. as Fig.
  3158. S2.
  3159. Do I need to do anything about that?
  3160. \end_layout
  3161. \end_inset
  3162. \end_layout
  3163. \begin_layout Plain Layout
  3164. \align center
  3165. \begin_inset Graphics
  3166. filename graphics/CD4-csaw/Promoter Peak Distance Profile-PAGE1-CROP.pdf
  3167. lyxscale 50
  3168. width 80col%
  3169. \end_inset
  3170. \end_layout
  3171. \begin_layout Plain Layout
  3172. \begin_inset Caption Standard
  3173. \begin_layout Plain Layout
  3174. \series bold
  3175. \begin_inset CommandInset label
  3176. LatexCommand label
  3177. name "fig:near-promoter-peak-enrich"
  3178. \end_inset
  3179. Enrichment of peaks in promoter neighborhoods.
  3180. \series default
  3181. This plot shows the distribution of distances from each annotated transcription
  3182. start site in the genome to the nearest called peak.
  3183. Each line represents one combination of histone mark, cell type, and time
  3184. point.
  3185. Distributions are smoothed using kernel density estimation [CITE? see ggplot2
  3186. stat_density()].
  3187. Transcription start sites that occur
  3188. \emph on
  3189. within
  3190. \emph default
  3191. peaks were excluded from this plot to avoid a large spike at zero that
  3192. would overshadow the rest of the distribution.
  3193. \end_layout
  3194. \end_inset
  3195. \end_layout
  3196. \end_inset
  3197. \end_layout
  3198. \begin_layout Standard
  3199. \begin_inset Float table
  3200. wide false
  3201. sideways false
  3202. status collapsed
  3203. \begin_layout Plain Layout
  3204. \align center
  3205. \begin_inset Tabular
  3206. <lyxtabular version="3" rows="4" columns="2">
  3207. <features tabularvalignment="middle">
  3208. <column alignment="center" valignment="top">
  3209. <column alignment="center" valignment="top">
  3210. <row>
  3211. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3212. \begin_inset Text
  3213. \begin_layout Plain Layout
  3214. Histone mark
  3215. \end_layout
  3216. \end_inset
  3217. </cell>
  3218. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3219. \begin_inset Text
  3220. \begin_layout Plain Layout
  3221. Effective promoter radius
  3222. \end_layout
  3223. \end_inset
  3224. </cell>
  3225. </row>
  3226. <row>
  3227. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3228. \begin_inset Text
  3229. \begin_layout Plain Layout
  3230. H3K4me2
  3231. \end_layout
  3232. \end_inset
  3233. </cell>
  3234. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3235. \begin_inset Text
  3236. \begin_layout Plain Layout
  3237. 1 kb
  3238. \end_layout
  3239. \end_inset
  3240. </cell>
  3241. </row>
  3242. <row>
  3243. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3244. \begin_inset Text
  3245. \begin_layout Plain Layout
  3246. H3K4me3
  3247. \end_layout
  3248. \end_inset
  3249. </cell>
  3250. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3251. \begin_inset Text
  3252. \begin_layout Plain Layout
  3253. 1 kb
  3254. \end_layout
  3255. \end_inset
  3256. </cell>
  3257. </row>
  3258. <row>
  3259. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3260. \begin_inset Text
  3261. \begin_layout Plain Layout
  3262. H3K27me3
  3263. \end_layout
  3264. \end_inset
  3265. </cell>
  3266. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3267. \begin_inset Text
  3268. \begin_layout Plain Layout
  3269. 2.5 kb
  3270. \end_layout
  3271. \end_inset
  3272. </cell>
  3273. </row>
  3274. </lyxtabular>
  3275. \end_inset
  3276. \end_layout
  3277. \begin_layout Plain Layout
  3278. \begin_inset Caption Standard
  3279. \begin_layout Plain Layout
  3280. \series bold
  3281. \begin_inset CommandInset label
  3282. LatexCommand label
  3283. name "tab:effective-promoter-radius"
  3284. \end_inset
  3285. Effective promoter radius for each histone mark.
  3286. \series default
  3287. These values represent the approximate distance from transcription start
  3288. site positions within which an excess of peaks are found, as shown in Figure
  3289. \begin_inset CommandInset ref
  3290. LatexCommand ref
  3291. reference "fig:near-promoter-peak-enrich"
  3292. plural "false"
  3293. caps "false"
  3294. noprefix "false"
  3295. \end_inset
  3296. .
  3297. \end_layout
  3298. \end_inset
  3299. \end_layout
  3300. \begin_layout Plain Layout
  3301. \end_layout
  3302. \end_inset
  3303. \end_layout
  3304. \begin_layout Standard
  3305. All 3 histone marks tend to occur more often near promoter regions, as shown
  3306. in Figure
  3307. \begin_inset CommandInset ref
  3308. LatexCommand ref
  3309. reference "fig:near-promoter-peak-enrich"
  3310. plural "false"
  3311. caps "false"
  3312. noprefix "false"
  3313. \end_inset
  3314. .
  3315. The majority of each density distribution is flat, representing the background
  3316. density of peaks genome-wide.
  3317. Each distribution has a peak near zero, representing an enrichment of peaks
  3318. close transcription start site (TSS) positions relative to the remainder
  3319. of the genome.
  3320. Interestingly, the
  3321. \begin_inset Quotes eld
  3322. \end_inset
  3323. radius
  3324. \begin_inset Quotes erd
  3325. \end_inset
  3326. within which this enrichment occurs is not the same for every histone mark
  3327. (Table
  3328. \begin_inset CommandInset ref
  3329. LatexCommand ref
  3330. reference "tab:effective-promoter-radius"
  3331. plural "false"
  3332. caps "false"
  3333. noprefix "false"
  3334. \end_inset
  3335. ).
  3336. For H3K4me2 and H3K4me3, peaks are most enriched within 1
  3337. \begin_inset space ~
  3338. \end_inset
  3339. kbp of TSS positions, while for H3K27me3, enrichment is broader, extending
  3340. to 2.5
  3341. \begin_inset space ~
  3342. \end_inset
  3343. kbp.
  3344. These
  3345. \begin_inset Quotes eld
  3346. \end_inset
  3347. effective promoter radii
  3348. \begin_inset Quotes erd
  3349. \end_inset
  3350. remain approximately the same across all combinations of experimental condition
  3351. (cell type, time point, and donor), so they appear to be a property of
  3352. the histone mark itself.
  3353. Hence, these radii were used to define the promoter regions for each histone
  3354. mark in all further analyses.
  3355. \end_layout
  3356. \begin_layout Standard
  3357. \begin_inset Flex TODO Note (inline)
  3358. status open
  3359. \begin_layout Plain Layout
  3360. Consider also showing figure for distance to nearest peak center, and reference
  3361. median peak size once that is known.
  3362. \end_layout
  3363. \end_inset
  3364. \end_layout
  3365. \begin_layout Subsection
  3366. H3K4 and H3K27 promoter methylation has broadly the expected correlation
  3367. with gene expression
  3368. \end_layout
  3369. \begin_layout Standard
  3370. \begin_inset Float figure
  3371. wide false
  3372. sideways false
  3373. status collapsed
  3374. \begin_layout Plain Layout
  3375. \begin_inset Flex TODO Note (inline)
  3376. status open
  3377. \begin_layout Plain Layout
  3378. This figure is generated from the old analysis.
  3379. Eiher note that in some way or re-generate it from the new peak calls.
  3380. \end_layout
  3381. \end_inset
  3382. \end_layout
  3383. \begin_layout Plain Layout
  3384. \align center
  3385. \begin_inset Graphics
  3386. filename graphics/CD4-csaw/FPKM by Peak Violin Plots-CROP.pdf
  3387. lyxscale 50
  3388. width 100col%
  3389. \end_inset
  3390. \end_layout
  3391. \begin_layout Plain Layout
  3392. \begin_inset Caption Standard
  3393. \begin_layout Plain Layout
  3394. \series bold
  3395. \begin_inset CommandInset label
  3396. LatexCommand label
  3397. name "fig:fpkm-by-peak"
  3398. \end_inset
  3399. Expression distributions of genes with and without promoter peaks.
  3400. \end_layout
  3401. \end_inset
  3402. \end_layout
  3403. \end_inset
  3404. \end_layout
  3405. \begin_layout Standard
  3406. H3K4me2 and H3K4me2 have previously been reported as activating marks whose
  3407. presence in a gene's promoter is associated with higher gene expression,
  3408. while H3K27me3 has been reported as inactivating [CITE].
  3409. The data are consistent with this characterization: genes whose promoters
  3410. (as defined by the radii for each histone mark listed in
  3411. \begin_inset CommandInset ref
  3412. LatexCommand ref
  3413. reference "tab:effective-promoter-radius"
  3414. plural "false"
  3415. caps "false"
  3416. noprefix "false"
  3417. \end_inset
  3418. ) overlap with a H3K4me2 or H3K4me3 peak tend to have higher expression
  3419. than those that don't, while H3K27me3 is likewise associated with lower
  3420. gene expression, as shown in
  3421. \begin_inset CommandInset ref
  3422. LatexCommand ref
  3423. reference "fig:fpkm-by-peak"
  3424. plural "false"
  3425. caps "false"
  3426. noprefix "false"
  3427. \end_inset
  3428. .
  3429. This pattern holds across all combinations of cell type and time point
  3430. (Welch's
  3431. \emph on
  3432. t
  3433. \emph default
  3434. -test, all
  3435. \begin_inset Formula $p\mathrm{-values}\ll2.2\times10^{-16}$
  3436. \end_inset
  3437. ).
  3438. The difference in average log FPKM values when a peak overlaps the promoter
  3439. is about
  3440. \begin_inset Formula $+5.67$
  3441. \end_inset
  3442. for H3K4me2,
  3443. \begin_inset Formula $+5.76$
  3444. \end_inset
  3445. for H3K4me2, and
  3446. \begin_inset Formula $-4.00$
  3447. \end_inset
  3448. for H3K27me3.
  3449. \end_layout
  3450. \begin_layout Standard
  3451. \begin_inset Flex TODO Note (inline)
  3452. status open
  3453. \begin_layout Plain Layout
  3454. I also have some figures looking at interactions between marks (e.g.
  3455. what if a promoter has both H3K4me3 and H3K27me3), but I don't know if
  3456. that much detail is warranted here, since all the effects just seem approximate
  3457. ly additive anyway.
  3458. \end_layout
  3459. \end_inset
  3460. \end_layout
  3461. \begin_layout Subsection
  3462. Gene expression and promoter histone methylation patterns in naive and memory
  3463. show convergence at day 14
  3464. \end_layout
  3465. \begin_layout Standard
  3466. \begin_inset ERT
  3467. status open
  3468. \begin_layout Plain Layout
  3469. \backslash
  3470. afterpage{
  3471. \end_layout
  3472. \begin_layout Plain Layout
  3473. \backslash
  3474. begin{landscape}
  3475. \end_layout
  3476. \end_inset
  3477. \end_layout
  3478. \begin_layout Standard
  3479. \begin_inset Float table
  3480. wide false
  3481. sideways false
  3482. status open
  3483. \begin_layout Plain Layout
  3484. \align center
  3485. \begin_inset Tabular
  3486. <lyxtabular version="3" rows="6" columns="7">
  3487. <features tabularvalignment="middle">
  3488. <column alignment="center" valignment="top">
  3489. <column alignment="center" valignment="top">
  3490. <column alignment="center" valignment="top">
  3491. <column alignment="center" valignment="top">
  3492. <column alignment="center" valignment="top">
  3493. <column alignment="center" valignment="top">
  3494. <column alignment="center" valignment="top">
  3495. <row>
  3496. <cell alignment="center" valignment="top" usebox="none">
  3497. \begin_inset Text
  3498. \begin_layout Plain Layout
  3499. \end_layout
  3500. \end_inset
  3501. </cell>
  3502. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3503. \begin_inset Text
  3504. \begin_layout Plain Layout
  3505. Number of significant promoters
  3506. \end_layout
  3507. \end_inset
  3508. </cell>
  3509. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3510. \begin_inset Text
  3511. \begin_layout Plain Layout
  3512. \end_layout
  3513. \end_inset
  3514. </cell>
  3515. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3516. \begin_inset Text
  3517. \begin_layout Plain Layout
  3518. \end_layout
  3519. \end_inset
  3520. </cell>
  3521. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3522. \begin_inset Text
  3523. \begin_layout Plain Layout
  3524. Est.
  3525. differentially modified promoters
  3526. \end_layout
  3527. \end_inset
  3528. </cell>
  3529. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3530. \begin_inset Text
  3531. \begin_layout Plain Layout
  3532. \end_layout
  3533. \end_inset
  3534. </cell>
  3535. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3536. \begin_inset Text
  3537. \begin_layout Plain Layout
  3538. \end_layout
  3539. \end_inset
  3540. </cell>
  3541. </row>
  3542. <row>
  3543. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3544. \begin_inset Text
  3545. \begin_layout Plain Layout
  3546. Time Point
  3547. \end_layout
  3548. \end_inset
  3549. </cell>
  3550. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3551. \begin_inset Text
  3552. \begin_layout Plain Layout
  3553. H3K4me2
  3554. \end_layout
  3555. \end_inset
  3556. </cell>
  3557. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3558. \begin_inset Text
  3559. \begin_layout Plain Layout
  3560. H3K4me3
  3561. \end_layout
  3562. \end_inset
  3563. </cell>
  3564. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3565. \begin_inset Text
  3566. \begin_layout Plain Layout
  3567. H3K27me3
  3568. \end_layout
  3569. \end_inset
  3570. </cell>
  3571. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3572. \begin_inset Text
  3573. \begin_layout Plain Layout
  3574. H3K4me2
  3575. \end_layout
  3576. \end_inset
  3577. </cell>
  3578. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3579. \begin_inset Text
  3580. \begin_layout Plain Layout
  3581. H3K4me3
  3582. \end_layout
  3583. \end_inset
  3584. </cell>
  3585. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3586. \begin_inset Text
  3587. \begin_layout Plain Layout
  3588. H3K27me3
  3589. \end_layout
  3590. \end_inset
  3591. </cell>
  3592. </row>
  3593. <row>
  3594. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3595. \begin_inset Text
  3596. \begin_layout Plain Layout
  3597. Day 0
  3598. \end_layout
  3599. \end_inset
  3600. </cell>
  3601. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3602. \begin_inset Text
  3603. \begin_layout Plain Layout
  3604. 4553
  3605. \end_layout
  3606. \end_inset
  3607. </cell>
  3608. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3609. \begin_inset Text
  3610. \begin_layout Plain Layout
  3611. 927
  3612. \end_layout
  3613. \end_inset
  3614. </cell>
  3615. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3616. \begin_inset Text
  3617. \begin_layout Plain Layout
  3618. 6
  3619. \end_layout
  3620. \end_inset
  3621. </cell>
  3622. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3623. \begin_inset Text
  3624. \begin_layout Plain Layout
  3625. 9967
  3626. \end_layout
  3627. \end_inset
  3628. </cell>
  3629. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3630. \begin_inset Text
  3631. \begin_layout Plain Layout
  3632. 4149
  3633. \end_layout
  3634. \end_inset
  3635. </cell>
  3636. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3637. \begin_inset Text
  3638. \begin_layout Plain Layout
  3639. 2404
  3640. \end_layout
  3641. \end_inset
  3642. </cell>
  3643. </row>
  3644. <row>
  3645. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3646. \begin_inset Text
  3647. \begin_layout Plain Layout
  3648. Day 1
  3649. \end_layout
  3650. \end_inset
  3651. </cell>
  3652. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3653. \begin_inset Text
  3654. \begin_layout Plain Layout
  3655. 567
  3656. \end_layout
  3657. \end_inset
  3658. </cell>
  3659. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3660. \begin_inset Text
  3661. \begin_layout Plain Layout
  3662. 278
  3663. \end_layout
  3664. \end_inset
  3665. </cell>
  3666. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3667. \begin_inset Text
  3668. \begin_layout Plain Layout
  3669. 1570
  3670. \end_layout
  3671. \end_inset
  3672. </cell>
  3673. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3674. \begin_inset Text
  3675. \begin_layout Plain Layout
  3676. 4370
  3677. \end_layout
  3678. \end_inset
  3679. </cell>
  3680. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3681. \begin_inset Text
  3682. \begin_layout Plain Layout
  3683. 2145
  3684. \end_layout
  3685. \end_inset
  3686. </cell>
  3687. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3688. \begin_inset Text
  3689. \begin_layout Plain Layout
  3690. 6598
  3691. \end_layout
  3692. \end_inset
  3693. </cell>
  3694. </row>
  3695. <row>
  3696. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3697. \begin_inset Text
  3698. \begin_layout Plain Layout
  3699. Day 5
  3700. \end_layout
  3701. \end_inset
  3702. </cell>
  3703. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3704. \begin_inset Text
  3705. \begin_layout Plain Layout
  3706. 2313
  3707. \end_layout
  3708. \end_inset
  3709. </cell>
  3710. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3711. \begin_inset Text
  3712. \begin_layout Plain Layout
  3713. 139
  3714. \end_layout
  3715. \end_inset
  3716. </cell>
  3717. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3718. \begin_inset Text
  3719. \begin_layout Plain Layout
  3720. 490
  3721. \end_layout
  3722. \end_inset
  3723. </cell>
  3724. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3725. \begin_inset Text
  3726. \begin_layout Plain Layout
  3727. 9450
  3728. \end_layout
  3729. \end_inset
  3730. </cell>
  3731. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3732. \begin_inset Text
  3733. \begin_layout Plain Layout
  3734. 1148
  3735. \end_layout
  3736. \end_inset
  3737. </cell>
  3738. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3739. \begin_inset Text
  3740. \begin_layout Plain Layout
  3741. 4141
  3742. \end_layout
  3743. \end_inset
  3744. </cell>
  3745. </row>
  3746. <row>
  3747. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3748. \begin_inset Text
  3749. \begin_layout Plain Layout
  3750. Day 14
  3751. \end_layout
  3752. \end_inset
  3753. </cell>
  3754. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3755. \begin_inset Text
  3756. \begin_layout Plain Layout
  3757. 0
  3758. \end_layout
  3759. \end_inset
  3760. </cell>
  3761. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3762. \begin_inset Text
  3763. \begin_layout Plain Layout
  3764. 0
  3765. \end_layout
  3766. \end_inset
  3767. </cell>
  3768. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3769. \begin_inset Text
  3770. \begin_layout Plain Layout
  3771. 0
  3772. \end_layout
  3773. \end_inset
  3774. </cell>
  3775. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3776. \begin_inset Text
  3777. \begin_layout Plain Layout
  3778. 0
  3779. \end_layout
  3780. \end_inset
  3781. </cell>
  3782. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3783. \begin_inset Text
  3784. \begin_layout Plain Layout
  3785. 0
  3786. \end_layout
  3787. \end_inset
  3788. </cell>
  3789. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3790. \begin_inset Text
  3791. \begin_layout Plain Layout
  3792. 0
  3793. \end_layout
  3794. \end_inset
  3795. </cell>
  3796. </row>
  3797. </lyxtabular>
  3798. \end_inset
  3799. \end_layout
  3800. \begin_layout Plain Layout
  3801. \begin_inset Caption Standard
  3802. \begin_layout Plain Layout
  3803. \series bold
  3804. \begin_inset CommandInset label
  3805. LatexCommand label
  3806. name "tab:Number-signif-promoters"
  3807. \end_inset
  3808. Number of differentially modified promoters between naive and memory cells
  3809. at each time point after activation.
  3810. \series default
  3811. This table shows both the number of differentially modified promoters detected
  3812. at a 10% FDR threshold (left half), and the total number of differentially
  3813. modified promoters as estimated using the method of
  3814. \begin_inset CommandInset citation
  3815. LatexCommand cite
  3816. key "Phipson2013"
  3817. literal "false"
  3818. \end_inset
  3819. (right half).
  3820. \end_layout
  3821. \end_inset
  3822. \end_layout
  3823. \end_inset
  3824. \end_layout
  3825. \begin_layout Standard
  3826. \begin_inset ERT
  3827. status open
  3828. \begin_layout Plain Layout
  3829. \backslash
  3830. end{landscape}
  3831. \end_layout
  3832. \begin_layout Plain Layout
  3833. }
  3834. \end_layout
  3835. \end_inset
  3836. \end_layout
  3837. \begin_layout Standard
  3838. \begin_inset Float figure
  3839. placement p
  3840. wide false
  3841. sideways false
  3842. status open
  3843. \begin_layout Plain Layout
  3844. \align center
  3845. \begin_inset Float figure
  3846. wide false
  3847. sideways false
  3848. status open
  3849. \begin_layout Plain Layout
  3850. \align center
  3851. \begin_inset Graphics
  3852. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-promoter-PCA-group-CROP.png
  3853. lyxscale 25
  3854. width 45col%
  3855. groupId pcoa-prom-subfig
  3856. \end_inset
  3857. \end_layout
  3858. \begin_layout Plain Layout
  3859. \begin_inset Caption Standard
  3860. \begin_layout Plain Layout
  3861. \series bold
  3862. \begin_inset CommandInset label
  3863. LatexCommand label
  3864. name "fig:PCoA-H3K4me2-prom"
  3865. \end_inset
  3866. PCoA plot of H3K4me2 promoters, after subtracting surrogate variables
  3867. \end_layout
  3868. \end_inset
  3869. \end_layout
  3870. \end_inset
  3871. \begin_inset space \hfill{}
  3872. \end_inset
  3873. \begin_inset Float figure
  3874. wide false
  3875. sideways false
  3876. status open
  3877. \begin_layout Plain Layout
  3878. \align center
  3879. \begin_inset Graphics
  3880. filename graphics/CD4-csaw/ChIP-seq/H3K4me3-promoter-PCA-group-CROP.png
  3881. lyxscale 25
  3882. width 45col%
  3883. groupId pcoa-prom-subfig
  3884. \end_inset
  3885. \end_layout
  3886. \begin_layout Plain Layout
  3887. \begin_inset Caption Standard
  3888. \begin_layout Plain Layout
  3889. \series bold
  3890. \begin_inset CommandInset label
  3891. LatexCommand label
  3892. name "fig:PCoA-H3K4me3-prom"
  3893. \end_inset
  3894. PCoA plot of H3K4me3 promoters, after subtracting surrogate variables
  3895. \end_layout
  3896. \end_inset
  3897. \end_layout
  3898. \end_inset
  3899. \end_layout
  3900. \begin_layout Plain Layout
  3901. \align center
  3902. \begin_inset Float figure
  3903. wide false
  3904. sideways false
  3905. status collapsed
  3906. \begin_layout Plain Layout
  3907. \align center
  3908. \begin_inset Graphics
  3909. filename graphics/CD4-csaw/ChIP-seq/H3K27me3-promoter-PCA-group-CROP.png
  3910. lyxscale 25
  3911. width 45col%
  3912. groupId pcoa-prom-subfig
  3913. \end_inset
  3914. \end_layout
  3915. \begin_layout Plain Layout
  3916. \begin_inset Caption Standard
  3917. \begin_layout Plain Layout
  3918. \series bold
  3919. \begin_inset CommandInset label
  3920. LatexCommand label
  3921. name "fig:PCoA-H3K27me3-prom"
  3922. \end_inset
  3923. PCoA plot of H3K27me3 promoters, after subtracting surrogate variables
  3924. \end_layout
  3925. \end_inset
  3926. \end_layout
  3927. \end_inset
  3928. \begin_inset space \hfill{}
  3929. \end_inset
  3930. \begin_inset Float figure
  3931. wide false
  3932. sideways false
  3933. status open
  3934. \begin_layout Plain Layout
  3935. \align center
  3936. \begin_inset Graphics
  3937. filename graphics/CD4-csaw/RNA-seq/PCA-final-23-CROP.png
  3938. lyxscale 25
  3939. width 45col%
  3940. groupId pcoa-prom-subfig
  3941. \end_inset
  3942. \end_layout
  3943. \begin_layout Plain Layout
  3944. \begin_inset Caption Standard
  3945. \begin_layout Plain Layout
  3946. \series bold
  3947. \begin_inset CommandInset label
  3948. LatexCommand label
  3949. name "fig:RNA-PCA-group"
  3950. \end_inset
  3951. RNA-seq PCoA showing principal coordiantes 2 and 3.
  3952. \end_layout
  3953. \end_inset
  3954. \end_layout
  3955. \end_inset
  3956. \end_layout
  3957. \begin_layout Plain Layout
  3958. \begin_inset Caption Standard
  3959. \begin_layout Plain Layout
  3960. \series bold
  3961. \begin_inset CommandInset label
  3962. LatexCommand label
  3963. name "fig:PCoA-promoters"
  3964. \end_inset
  3965. PCoA plots for promoter ChIP-seq and expression RNA-seq data
  3966. \end_layout
  3967. \end_inset
  3968. \end_layout
  3969. \end_inset
  3970. \end_layout
  3971. \begin_layout Standard
  3972. \begin_inset Flex TODO Note (inline)
  3973. status open
  3974. \begin_layout Plain Layout
  3975. Check up on figure refs in this paragraph
  3976. \end_layout
  3977. \end_inset
  3978. \end_layout
  3979. \begin_layout Standard
  3980. We hypothesized that if naive cells had differentiated into memory cells
  3981. by Day 14, then their patterns of expression and histone modification should
  3982. converge with those of memory cells at Day 14.
  3983. Figure
  3984. \begin_inset CommandInset ref
  3985. LatexCommand ref
  3986. reference "fig:PCoA-promoters"
  3987. plural "false"
  3988. caps "false"
  3989. noprefix "false"
  3990. \end_inset
  3991. shows the patterns of variation in all 3 histone marks in the promoter
  3992. regions of the genome using principal coordinate analysis.
  3993. All 3 marks show a noticeable convergence between the naive and memory
  3994. samples at day 14, visible as an overlapping of the day 14 groups on each
  3995. plot.
  3996. This is consistent with the counts of significantly differentially modified
  3997. promoters and estimates of the total numbers of differentially modified
  3998. promoters shown in Table
  3999. \begin_inset CommandInset ref
  4000. LatexCommand ref
  4001. reference "tab:Number-signif-promoters"
  4002. plural "false"
  4003. caps "false"
  4004. noprefix "false"
  4005. \end_inset
  4006. .
  4007. For all histone marks, evidence of differential modification between naive
  4008. and memory samples was detected at every time point except day 14.
  4009. The day 14 convergence pattern is also present in the RNA-seq data (Figure
  4010. \begin_inset CommandInset ref
  4011. LatexCommand ref
  4012. reference "fig:RNA-PCA-group"
  4013. plural "false"
  4014. caps "false"
  4015. noprefix "false"
  4016. \end_inset
  4017. ), albiet in the 2nd and 3rd principal coordinates, indicating that it is
  4018. not the most dominant pattern driving gene expression.
  4019. Taken together, the data show that promoter histone methylation for these
  4020. 3 histone marks and RNA expression for naive and memory cells are most
  4021. similar at day 14, the furthest time point after activation.
  4022. MOFA was also able to capture this day 14 convergence pattern in latent
  4023. factor 5 (Figure
  4024. \begin_inset CommandInset ref
  4025. LatexCommand ref
  4026. reference "fig:mofa-lf-scatter"
  4027. plural "false"
  4028. caps "false"
  4029. noprefix "false"
  4030. \end_inset
  4031. ), which accounts for shared variation across all 3 histone marks and the
  4032. RNA-seq data, confirming that this convergence is a coordinated pattern
  4033. across all 4 data sets.
  4034. While this observation does not prove that the naive cells have differentiated
  4035. into memory cells at Day 14, it is consistent with that hypothesis.
  4036. \end_layout
  4037. \begin_layout Subsection
  4038. Effect of H3K4me2 and H3K4me3 promoter coverage upstream vs downstream of
  4039. TSS
  4040. \end_layout
  4041. \begin_layout Standard
  4042. \begin_inset Flex TODO Note (inline)
  4043. status open
  4044. \begin_layout Plain Layout
  4045. Need a better section title, for this and the next one.
  4046. \end_layout
  4047. \end_inset
  4048. \end_layout
  4049. \begin_layout Standard
  4050. \begin_inset Flex TODO Note (inline)
  4051. status open
  4052. \begin_layout Plain Layout
  4053. Make sure use of coverage/abundance/whatever is consistent.
  4054. \end_layout
  4055. \end_inset
  4056. \end_layout
  4057. \begin_layout Standard
  4058. \begin_inset Flex TODO Note (inline)
  4059. status open
  4060. \begin_layout Plain Layout
  4061. For the figures in this section and the next, the group labels are arbitrary,
  4062. so if time allows, it would be good to manually reorder them in a logical
  4063. way, e.g.
  4064. most upstream to most downstream.
  4065. If this is done, make sure to update the text with the correct group labels.
  4066. \end_layout
  4067. \end_inset
  4068. \end_layout
  4069. \begin_layout Standard
  4070. \begin_inset ERT
  4071. status open
  4072. \begin_layout Plain Layout
  4073. \backslash
  4074. afterpage{
  4075. \end_layout
  4076. \begin_layout Plain Layout
  4077. \backslash
  4078. begin{landscape}
  4079. \end_layout
  4080. \end_inset
  4081. \end_layout
  4082. \begin_layout Standard
  4083. \begin_inset Float figure
  4084. wide false
  4085. sideways false
  4086. status open
  4087. \begin_layout Plain Layout
  4088. \align center
  4089. \begin_inset Float figure
  4090. wide false
  4091. sideways false
  4092. status open
  4093. \begin_layout Plain Layout
  4094. \align center
  4095. \begin_inset Graphics
  4096. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-neighborhood-clusters-CROP.png
  4097. lyxscale 25
  4098. width 30col%
  4099. groupId covprof-subfig
  4100. \end_inset
  4101. \end_layout
  4102. \begin_layout Plain Layout
  4103. \begin_inset Caption Standard
  4104. \begin_layout Plain Layout
  4105. \series bold
  4106. \begin_inset CommandInset label
  4107. LatexCommand label
  4108. name "fig:H3K4me2-neighborhood-clusters"
  4109. \end_inset
  4110. Average relative coverage for each bin in each cluster
  4111. \end_layout
  4112. \end_inset
  4113. \end_layout
  4114. \end_inset
  4115. \begin_inset space \hfill{}
  4116. \end_inset
  4117. \begin_inset Float figure
  4118. wide false
  4119. sideways false
  4120. status open
  4121. \begin_layout Plain Layout
  4122. \align center
  4123. \begin_inset Graphics
  4124. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-neighborhood-PCA-CROP.png
  4125. lyxscale 25
  4126. width 30col%
  4127. groupId covprof-subfig
  4128. \end_inset
  4129. \end_layout
  4130. \begin_layout Plain Layout
  4131. \begin_inset Caption Standard
  4132. \begin_layout Plain Layout
  4133. \series bold
  4134. \begin_inset CommandInset label
  4135. LatexCommand label
  4136. name "fig:H3K4me2-neighborhood-pca"
  4137. \end_inset
  4138. PCA of relative coverage depth, colored by K-means cluster membership.
  4139. \end_layout
  4140. \end_inset
  4141. \end_layout
  4142. \end_inset
  4143. \begin_inset space \hfill{}
  4144. \end_inset
  4145. \begin_inset Float figure
  4146. wide false
  4147. sideways false
  4148. status open
  4149. \begin_layout Plain Layout
  4150. \align center
  4151. \begin_inset Graphics
  4152. filename graphics/CD4-csaw/ChIP-seq/H3K4me2-neighborhood-expression-CROP.png
  4153. lyxscale 25
  4154. width 30col%
  4155. groupId covprof-subfig
  4156. \end_inset
  4157. \end_layout
  4158. \begin_layout Plain Layout
  4159. \begin_inset Caption Standard
  4160. \begin_layout Plain Layout
  4161. \series bold
  4162. \begin_inset CommandInset label
  4163. LatexCommand label
  4164. name "fig:H3K4me2-neighborhood-expression"
  4165. \end_inset
  4166. Gene expression grouped by promoter coverage clusters.
  4167. \end_layout
  4168. \end_inset
  4169. \end_layout
  4170. \end_inset
  4171. \end_layout
  4172. \begin_layout Plain Layout
  4173. \begin_inset Caption Standard
  4174. \begin_layout Plain Layout
  4175. \series bold
  4176. \begin_inset CommandInset label
  4177. LatexCommand label
  4178. name "fig:H3K4me2-neighborhood"
  4179. \end_inset
  4180. K-means clustering of promoter H3K4me2 relative coverage depth in naive
  4181. day 0 samples.
  4182. \series default
  4183. H3K4me2 ChIP-seq reads were binned into 500-bp windows tiled across each
  4184. promoter from 5
  4185. \begin_inset space ~
  4186. \end_inset
  4187. kbp upstream to 5
  4188. \begin_inset space ~
  4189. \end_inset
  4190. kbp downstream, and the logCPM values were normalized within each promoter
  4191. to an average of 0, yielding relative coverage depths.
  4192. These were then grouped using K-means clustering with
  4193. \begin_inset Formula $K=6$
  4194. \end_inset
  4195. ,
  4196. \series bold
  4197. \series default
  4198. and the average bin values were plotted for each cluster (a).
  4199. The
  4200. \begin_inset Formula $x$
  4201. \end_inset
  4202. -axis is the genomic coordinate of each bin relative to the the transcription
  4203. start site, and the
  4204. \begin_inset Formula $y$
  4205. \end_inset
  4206. -axis is the mean relative coverage depth of that bin across all promoters
  4207. in the cluster.
  4208. Each line represents the average
  4209. \begin_inset Quotes eld
  4210. \end_inset
  4211. shape
  4212. \begin_inset Quotes erd
  4213. \end_inset
  4214. of the promoter coverage for promoters in that cluster.
  4215. PCA was performed on the same data, and the first two principal components
  4216. were plotted, coloring each point by its K-means cluster identity (b).
  4217. For each cluster, the distribution of gene expression values was plotted
  4218. (c).
  4219. \end_layout
  4220. \end_inset
  4221. \end_layout
  4222. \end_inset
  4223. \end_layout
  4224. \begin_layout Standard
  4225. \begin_inset ERT
  4226. status open
  4227. \begin_layout Plain Layout
  4228. \backslash
  4229. end{landscape}
  4230. \end_layout
  4231. \begin_layout Plain Layout
  4232. }
  4233. \end_layout
  4234. \end_inset
  4235. \end_layout
  4236. \begin_layout Standard
  4237. To test whether the position of a histone mark relative to a gene's transcriptio
  4238. n start site (TSS) was important, we looked at the
  4239. \begin_inset Quotes eld
  4240. \end_inset
  4241. landscape
  4242. \begin_inset Quotes erd
  4243. \end_inset
  4244. of ChIP-seq read coverage in naive Day 0 samples within 5 kb of each gene's
  4245. TSS by binning reads into 500-bp windows tiled across each promoter LogCPM
  4246. values were calculated for the bins in each promoter and then the average
  4247. logCPM for each promoter's bins was normalized to zero, such that the values
  4248. represent coverage relative to other regions of the same promoter rather
  4249. than being proportional to absolute read count.
  4250. The promoters were then clustered based on the normalized bin abundances
  4251. using
  4252. \begin_inset Formula $k$
  4253. \end_inset
  4254. -means clustering with
  4255. \begin_inset Formula $K=6$
  4256. \end_inset
  4257. .
  4258. Different values of
  4259. \begin_inset Formula $K$
  4260. \end_inset
  4261. were also tested, but did not substantially change the interpretation of
  4262. the data.
  4263. \end_layout
  4264. \begin_layout Standard
  4265. For H3K4me2, plotting the average bin abundances for each cluster reveals
  4266. a simple pattern (Figure
  4267. \begin_inset CommandInset ref
  4268. LatexCommand ref
  4269. reference "fig:H3K4me2-neighborhood-clusters"
  4270. plural "false"
  4271. caps "false"
  4272. noprefix "false"
  4273. \end_inset
  4274. ): Cluster 5 represents a completely flat promoter coverage profile, likely
  4275. consisting of genes with no H3K4me2 methylation in the promoter.
  4276. All the other clusters represent a continuum of peak positions relative
  4277. to the TSS.
  4278. In order from must upstream to most downstream, they are Clusters 6, 4,
  4279. 3, 1, and 2.
  4280. There do not appear to be any clusters representing coverage patterns other
  4281. than lone peaks, such as coverage troughs or double peaks.
  4282. Next, all promoters were plotted in a PCA plot based on the same relative
  4283. bin abundance data, and colored based on cluster membership (Figure
  4284. \begin_inset CommandInset ref
  4285. LatexCommand ref
  4286. reference "fig:H3K4me2-neighborhood-pca"
  4287. plural "false"
  4288. caps "false"
  4289. noprefix "false"
  4290. \end_inset
  4291. ).
  4292. The PCA plot shows Cluster 5 (the
  4293. \begin_inset Quotes eld
  4294. \end_inset
  4295. no peak
  4296. \begin_inset Quotes erd
  4297. \end_inset
  4298. cluster) at the center, with the other clusters arranged in a counter-clockwise
  4299. arc around it in the order noted above, from most upstream peak to most
  4300. downstream.
  4301. Notably, the
  4302. \begin_inset Quotes eld
  4303. \end_inset
  4304. clusters
  4305. \begin_inset Quotes erd
  4306. \end_inset
  4307. form a single large
  4308. \begin_inset Quotes eld
  4309. \end_inset
  4310. cloud
  4311. \begin_inset Quotes erd
  4312. \end_inset
  4313. with no apparent separation between them, further supporting the conclusion
  4314. that these clusters represent an arbitrary partitioning of a continuous
  4315. distribution of promoter coverage landscapes.
  4316. While the clusters are a useful abstraction that aids in visualization,
  4317. they are ultimately not an accurate representation of the data.
  4318. A better representation might be something like a polar coordinate system
  4319. with the origin at the center of Cluster 5, where the radius represents
  4320. the peak height above the background and the angle represents the peak's
  4321. position upstream or downstream of the TSS.
  4322. The continuous nature of the distribution also explains why different values
  4323. of
  4324. \begin_inset Formula $K$
  4325. \end_inset
  4326. led to similar conclusions.
  4327. \end_layout
  4328. \begin_layout Standard
  4329. \begin_inset Flex TODO Note (inline)
  4330. status open
  4331. \begin_layout Plain Layout
  4332. RNA-seq values in the plots use logCPM but should really use logFPKM or
  4333. logTPM.
  4334. Fix if time allows.
  4335. \end_layout
  4336. \end_inset
  4337. \end_layout
  4338. \begin_layout Standard
  4339. \begin_inset Flex TODO Note (inline)
  4340. status open
  4341. \begin_layout Plain Layout
  4342. Should have a table of p-values on difference of means between Cluster 5
  4343. and the others.
  4344. \end_layout
  4345. \end_inset
  4346. \end_layout
  4347. \begin_layout Standard
  4348. To investigate the association between relative peak position and gene expressio
  4349. n, we plotted the Naive Day 0 expression for the genes in each cluster (Figure
  4350. \begin_inset CommandInset ref
  4351. LatexCommand ref
  4352. reference "fig:H3K4me2-neighborhood-expression"
  4353. plural "false"
  4354. caps "false"
  4355. noprefix "false"
  4356. \end_inset
  4357. ).
  4358. Most genes in Cluster 5, the
  4359. \begin_inset Quotes eld
  4360. \end_inset
  4361. no peak
  4362. \begin_inset Quotes erd
  4363. \end_inset
  4364. cluster, have low expression values.
  4365. Taking this as the
  4366. \begin_inset Quotes eld
  4367. \end_inset
  4368. baseline
  4369. \begin_inset Quotes erd
  4370. \end_inset
  4371. distribution when no H3K4me2 methylation is present, we can compare the
  4372. other clusters' distributions to determine which peak positions are associated
  4373. with elevated expression.
  4374. As might be expected, the 3 clusters representing peaks closest to the
  4375. TSS, Clusters 1, 3, and 4, show the highest average expression distributions.
  4376. Specifically, these clusters all have their highest ChIP-seq abundance
  4377. within 1kb of the TSS, consistent with the previously determined promoter
  4378. radius.
  4379. In contrast, cluster 6, which represents peaks several kb upstream of the
  4380. TSS, shows a slightly higher average expression than baseline, while Cluster
  4381. 2, which represents peaks several kb downstream, doesn't appear to show
  4382. any appreciable difference.
  4383. Interestingly, the cluster with the highest average expression is Cluster
  4384. 1, which represents peaks about 1 kb downstream of the TSS, rather than
  4385. Cluster 3, which represents peaks centered directly at the TSS.
  4386. This suggests that conceptualizing the promoter as a region centered on
  4387. the TSS with a certain
  4388. \begin_inset Quotes eld
  4389. \end_inset
  4390. radius
  4391. \begin_inset Quotes erd
  4392. \end_inset
  4393. may be an oversimplification – a peak that is a specific distance from
  4394. the TSS may have a different degree of influence depending on whether it
  4395. is upstream or downstream of the TSS.
  4396. \end_layout
  4397. \begin_layout Standard
  4398. \begin_inset ERT
  4399. status open
  4400. \begin_layout Plain Layout
  4401. \backslash
  4402. afterpage{
  4403. \end_layout
  4404. \begin_layout Plain Layout
  4405. \backslash
  4406. begin{landscape}
  4407. \end_layout
  4408. \end_inset
  4409. \end_layout
  4410. \begin_layout Standard
  4411. \begin_inset Float figure
  4412. wide false
  4413. sideways false
  4414. status open
  4415. \begin_layout Plain Layout
  4416. \align center
  4417. \begin_inset Float figure
  4418. wide false
  4419. sideways false
  4420. status open
  4421. \begin_layout Plain Layout
  4422. \align center
  4423. \begin_inset Graphics
  4424. filename graphics/CD4-csaw/ChIP-seq/H3K4me3-neighborhood-clusters-CROP.png
  4425. lyxscale 25
  4426. width 30col%
  4427. groupId covprof-subfig
  4428. \end_inset
  4429. \end_layout
  4430. \begin_layout Plain Layout
  4431. \begin_inset Caption Standard
  4432. \begin_layout Plain Layout
  4433. \series bold
  4434. \begin_inset CommandInset label
  4435. LatexCommand label
  4436. name "fig:H3K4me3-neighborhood-clusters"
  4437. \end_inset
  4438. Average relative coverage for each bin in each cluster
  4439. \end_layout
  4440. \end_inset
  4441. \end_layout
  4442. \end_inset
  4443. \begin_inset space \hfill{}
  4444. \end_inset
  4445. \begin_inset Float figure
  4446. wide false
  4447. sideways false
  4448. status open
  4449. \begin_layout Plain Layout
  4450. \align center
  4451. \begin_inset Graphics
  4452. filename graphics/CD4-csaw/ChIP-seq/H3K4me3-neighborhood-PCA-CROP.png
  4453. lyxscale 25
  4454. width 30col%
  4455. groupId covprof-subfig
  4456. \end_inset
  4457. \end_layout
  4458. \begin_layout Plain Layout
  4459. \begin_inset Caption Standard
  4460. \begin_layout Plain Layout
  4461. \series bold
  4462. \begin_inset CommandInset label
  4463. LatexCommand label
  4464. name "fig:H3K4me3-neighborhood-pca"
  4465. \end_inset
  4466. PCA of relative coverage depth, colored by K-means cluster membership.
  4467. \end_layout
  4468. \end_inset
  4469. \end_layout
  4470. \end_inset
  4471. \begin_inset space \hfill{}
  4472. \end_inset
  4473. \begin_inset Float figure
  4474. wide false
  4475. sideways false
  4476. status open
  4477. \begin_layout Plain Layout
  4478. \align center
  4479. \begin_inset Graphics
  4480. filename graphics/CD4-csaw/ChIP-seq/H3K4me3-neighborhood-expression-CROP.png
  4481. lyxscale 25
  4482. width 30col%
  4483. groupId covprof-subfig
  4484. \end_inset
  4485. \end_layout
  4486. \begin_layout Plain Layout
  4487. \begin_inset Caption Standard
  4488. \begin_layout Plain Layout
  4489. \series bold
  4490. \begin_inset CommandInset label
  4491. LatexCommand label
  4492. name "fig:H3K4me3-neighborhood-expression"
  4493. \end_inset
  4494. Gene expression grouped by promoter coverage clusters.
  4495. \end_layout
  4496. \end_inset
  4497. \end_layout
  4498. \end_inset
  4499. \end_layout
  4500. \begin_layout Plain Layout
  4501. \begin_inset Caption Standard
  4502. \begin_layout Plain Layout
  4503. \series bold
  4504. \begin_inset CommandInset label
  4505. LatexCommand label
  4506. name "fig:H3K4me3-neighborhood"
  4507. \end_inset
  4508. K-means clustering of promoter H3K4me3 relative coverage depth in naive
  4509. day 0 samples.
  4510. \series default
  4511. H3K4me2 ChIP-seq reads were binned into 500-bp windows tiled across each
  4512. promoter from 5
  4513. \begin_inset space ~
  4514. \end_inset
  4515. kbp upstream to 5
  4516. \begin_inset space ~
  4517. \end_inset
  4518. kbp downstream, and the logCPM values were normalized within each promoter
  4519. to an average of 0, yielding relative coverage depths.
  4520. These were then grouped using K-means clustering with
  4521. \begin_inset Formula $K=6$
  4522. \end_inset
  4523. ,
  4524. \series bold
  4525. \series default
  4526. and the average bin values were plotted for each cluster (a).
  4527. The
  4528. \begin_inset Formula $x$
  4529. \end_inset
  4530. -axis is the genomic coordinate of each bin relative to the the transcription
  4531. start site, and the
  4532. \begin_inset Formula $y$
  4533. \end_inset
  4534. -axis is the mean relative coverage depth of that bin across all promoters
  4535. in the cluster.
  4536. Each line represents the average
  4537. \begin_inset Quotes eld
  4538. \end_inset
  4539. shape
  4540. \begin_inset Quotes erd
  4541. \end_inset
  4542. of the promoter coverage for promoters in that cluster.
  4543. PCA was performed on the same data, and the first two principal components
  4544. were plotted, coloring each point by its K-means cluster identity (b).
  4545. For each cluster, the distribution of gene expression values was plotted
  4546. (c).
  4547. \end_layout
  4548. \end_inset
  4549. \end_layout
  4550. \end_inset
  4551. \end_layout
  4552. \begin_layout Standard
  4553. \begin_inset ERT
  4554. status open
  4555. \begin_layout Plain Layout
  4556. \backslash
  4557. end{landscape}
  4558. \end_layout
  4559. \begin_layout Plain Layout
  4560. }
  4561. \end_layout
  4562. \end_inset
  4563. \end_layout
  4564. \begin_layout Standard
  4565. \begin_inset Flex TODO Note (inline)
  4566. status open
  4567. \begin_layout Plain Layout
  4568. Is there more to say here?
  4569. \end_layout
  4570. \end_inset
  4571. \end_layout
  4572. \begin_layout Standard
  4573. All observations described above for H3K4me2 ChIP-seq also appear to hold
  4574. for H3K4me3 as well (Figure
  4575. \begin_inset CommandInset ref
  4576. LatexCommand ref
  4577. reference "fig:H3K4me3-neighborhood"
  4578. plural "false"
  4579. caps "false"
  4580. noprefix "false"
  4581. \end_inset
  4582. ).
  4583. This is expected, since there is a high correlation between the positions
  4584. where both histone marks occur.
  4585. \end_layout
  4586. \begin_layout Subsection
  4587. Promoter coverage H3K27me3
  4588. \end_layout
  4589. \begin_layout Standard
  4590. \begin_inset ERT
  4591. status open
  4592. \begin_layout Plain Layout
  4593. \backslash
  4594. afterpage{
  4595. \end_layout
  4596. \begin_layout Plain Layout
  4597. \backslash
  4598. begin{landscape}
  4599. \end_layout
  4600. \end_inset
  4601. \end_layout
  4602. \begin_layout Standard
  4603. \begin_inset Float figure
  4604. wide false
  4605. sideways false
  4606. status collapsed
  4607. \begin_layout Plain Layout
  4608. \align center
  4609. \begin_inset Float figure
  4610. wide false
  4611. sideways false
  4612. status open
  4613. \begin_layout Plain Layout
  4614. \align center
  4615. \begin_inset Graphics
  4616. filename graphics/CD4-csaw/ChIP-seq/H3K27me3-neighborhood-clusters-CROP.png
  4617. lyxscale 25
  4618. width 30col%
  4619. groupId covprof-subfig
  4620. \end_inset
  4621. \end_layout
  4622. \begin_layout Plain Layout
  4623. \begin_inset Caption Standard
  4624. \begin_layout Plain Layout
  4625. \series bold
  4626. \begin_inset CommandInset label
  4627. LatexCommand label
  4628. name "fig:H3K27me3-neighborhood-clusters"
  4629. \end_inset
  4630. Average relative coverage for each bin in each cluster
  4631. \end_layout
  4632. \end_inset
  4633. \end_layout
  4634. \end_inset
  4635. \begin_inset space \hfill{}
  4636. \end_inset
  4637. \begin_inset Float figure
  4638. wide false
  4639. sideways false
  4640. status open
  4641. \begin_layout Plain Layout
  4642. \align center
  4643. \begin_inset Graphics
  4644. filename graphics/CD4-csaw/ChIP-seq/H3K27me3-neighborhood-PCA-CROP.png
  4645. lyxscale 25
  4646. width 30col%
  4647. groupId covprof-subfig
  4648. \end_inset
  4649. \end_layout
  4650. \begin_layout Plain Layout
  4651. \begin_inset Caption Standard
  4652. \begin_layout Plain Layout
  4653. \series bold
  4654. \begin_inset CommandInset label
  4655. LatexCommand label
  4656. name "fig:H3K27me3-neighborhood-pca"
  4657. \end_inset
  4658. PCA of relative coverage depth, colored by K-means cluster membership.
  4659. \series default
  4660. Note that Cluster 6 is hidden behind all the other clusters.
  4661. \end_layout
  4662. \end_inset
  4663. \end_layout
  4664. \end_inset
  4665. \begin_inset space \hfill{}
  4666. \end_inset
  4667. \begin_inset Float figure
  4668. wide false
  4669. sideways false
  4670. status open
  4671. \begin_layout Plain Layout
  4672. \align center
  4673. \begin_inset Graphics
  4674. filename graphics/CD4-csaw/ChIP-seq/H3K27me3-neighborhood-expression-CROP.png
  4675. lyxscale 25
  4676. width 30col%
  4677. groupId covprof-subfig
  4678. \end_inset
  4679. \end_layout
  4680. \begin_layout Plain Layout
  4681. \begin_inset Caption Standard
  4682. \begin_layout Plain Layout
  4683. \series bold
  4684. \begin_inset CommandInset label
  4685. LatexCommand label
  4686. name "fig:H3K27me3-neighborhood-expression"
  4687. \end_inset
  4688. Gene expression grouped by promoter coverage clusters.
  4689. \end_layout
  4690. \end_inset
  4691. \end_layout
  4692. \end_inset
  4693. \end_layout
  4694. \begin_layout Plain Layout
  4695. \begin_inset Flex TODO Note (inline)
  4696. status open
  4697. \begin_layout Plain Layout
  4698. Repeated figure legends are kind of an issue here.
  4699. What to do?
  4700. \end_layout
  4701. \end_inset
  4702. \end_layout
  4703. \begin_layout Plain Layout
  4704. \begin_inset Caption Standard
  4705. \begin_layout Plain Layout
  4706. \series bold
  4707. \begin_inset CommandInset label
  4708. LatexCommand label
  4709. name "fig:H3K27me3-neighborhood"
  4710. \end_inset
  4711. K-means clustering of promoter H3K27me3 relative coverage depth in naive
  4712. day 0 samples.
  4713. \series default
  4714. H3K27me3 ChIP-seq reads were binned into 500-bp windows tiled across each
  4715. promoter from 5
  4716. \begin_inset space ~
  4717. \end_inset
  4718. kbp upstream to 5
  4719. \begin_inset space ~
  4720. \end_inset
  4721. kbp downstream, and the logCPM values were normalized within each promoter
  4722. to an average of 0, yielding relative coverage depths.
  4723. These were then grouped using
  4724. \begin_inset Formula $k$
  4725. \end_inset
  4726. -means clustering with
  4727. \begin_inset Formula $K=6$
  4728. \end_inset
  4729. ,
  4730. \series bold
  4731. \series default
  4732. and the average bin values were plotted for each cluster (a).
  4733. The
  4734. \begin_inset Formula $x$
  4735. \end_inset
  4736. -axis is the genomic coordinate of each bin relative to the the transcription
  4737. start site, and the
  4738. \begin_inset Formula $y$
  4739. \end_inset
  4740. -axis is the mean relative coverage depth of that bin across all promoters
  4741. in the cluster.
  4742. Each line represents the average
  4743. \begin_inset Quotes eld
  4744. \end_inset
  4745. shape
  4746. \begin_inset Quotes erd
  4747. \end_inset
  4748. of the promoter coverage for promoters in that cluster.
  4749. PCA was performed on the same data, and the first two principal components
  4750. were plotted, coloring each point by its K-means cluster identity (b).
  4751. For each cluster, the distribution of gene expression values was plotted
  4752. (c).
  4753. \end_layout
  4754. \end_inset
  4755. \end_layout
  4756. \end_inset
  4757. \end_layout
  4758. \begin_layout Standard
  4759. \begin_inset ERT
  4760. status open
  4761. \begin_layout Plain Layout
  4762. \backslash
  4763. end{landscape}
  4764. \end_layout
  4765. \begin_layout Plain Layout
  4766. }
  4767. \end_layout
  4768. \end_inset
  4769. \end_layout
  4770. \begin_layout Standard
  4771. \begin_inset Flex TODO Note (inline)
  4772. status open
  4773. \begin_layout Plain Layout
  4774. Should maybe re-explain what was done or refer back to the previous section.
  4775. \end_layout
  4776. \end_inset
  4777. \end_layout
  4778. \begin_layout Standard
  4779. Unlike both H3K4 marks, whose main patterns of variation appear directly
  4780. related to the size and position of a single peak within the promoter,
  4781. the patterns of H3K27me3 methylation in promoters are more complex (Figure
  4782. \begin_inset CommandInset ref
  4783. LatexCommand ref
  4784. reference "fig:H3K27me3-neighborhood"
  4785. plural "false"
  4786. caps "false"
  4787. noprefix "false"
  4788. \end_inset
  4789. ).
  4790. Once again looking at the relative coverage in a 500-bp wide bins in a
  4791. 5kb radius around each TSS, promoters were clustered based on the normalized
  4792. relative coverage values in each bin using
  4793. \begin_inset Formula $k$
  4794. \end_inset
  4795. -means clustering with
  4796. \begin_inset Formula $K=6$
  4797. \end_inset
  4798. (Figure
  4799. \begin_inset CommandInset ref
  4800. LatexCommand ref
  4801. reference "fig:H3K27me3-neighborhood-clusters"
  4802. plural "false"
  4803. caps "false"
  4804. noprefix "false"
  4805. \end_inset
  4806. ).
  4807. This time, 3
  4808. \begin_inset Quotes eld
  4809. \end_inset
  4810. axes
  4811. \begin_inset Quotes erd
  4812. \end_inset
  4813. of variation can be observed, each represented by 2 clusters with opposing
  4814. patterns.
  4815. The first axis is greater upstream coverage (Cluster 1) vs.
  4816. greater downstream coverage (Cluster 3); the second axis is the coverage
  4817. at the TSS itself: peak (Cluster 4) or trough (Cluster 2); lastly, the
  4818. third axis represents a trough upstream of the TSS (Cluster 5) vs.
  4819. downstream of the TSS (Cluster 6).
  4820. Referring to these opposing pairs of clusters as axes of variation is justified
  4821. , because they correspond precisely to the first 3 principal components
  4822. in the PCA plot of the relative coverage values (Figure
  4823. \begin_inset CommandInset ref
  4824. LatexCommand ref
  4825. reference "fig:H3K27me3-neighborhood-pca"
  4826. plural "false"
  4827. caps "false"
  4828. noprefix "false"
  4829. \end_inset
  4830. ).
  4831. The PCA plot reveals that as in the case of H3K4me2, all the
  4832. \begin_inset Quotes eld
  4833. \end_inset
  4834. clusters
  4835. \begin_inset Quotes erd
  4836. \end_inset
  4837. are really just sections of a single connected cloud rather than discrete
  4838. clusters.
  4839. The cloud is approximately ellipsoid-shaped, with each PC being an axis
  4840. of the ellipse, and each cluster consisting of a pyrimidal section of the
  4841. ellipsoid.
  4842. \end_layout
  4843. \begin_layout Standard
  4844. In Figure
  4845. \begin_inset CommandInset ref
  4846. LatexCommand ref
  4847. reference "fig:H3K27me3-neighborhood-expression"
  4848. plural "false"
  4849. caps "false"
  4850. noprefix "false"
  4851. \end_inset
  4852. , we can see that Clusters 1 and 2 are the only clusters with higher gene
  4853. expression than the others.
  4854. For Cluster 2, this is expected, since this cluster represents genes with
  4855. depletion of H3K27me3 near the promoter.
  4856. Hence, elevated expression in cluster 2 is consistent with the conventional
  4857. view of H3K27me3 as a deactivating mark.
  4858. However, Cluster 1, the cluster with the most elevated gene expression,
  4859. represents genes with elevated coverage upstream of the TSS, or equivalently,
  4860. decreased coverage downstream, inside the gene body.
  4861. The opposite pattern, in which H3K27me3 is more abundant within the gene
  4862. body and less abundance in the upstream promoter region, does not show
  4863. any elevation in gene expression.
  4864. As with H3K4me2, this shows that the location of H3K27 trimethylation relative
  4865. to the TSS is potentially an important factor beyond simple proximity.
  4866. \end_layout
  4867. \begin_layout Standard
  4868. \begin_inset Flex TODO Note (inline)
  4869. status open
  4870. \begin_layout Plain Layout
  4871. Show the figures where the negative result ended this line of inquiry.
  4872. I need to debug some errors resulting from an R upgrade to do this.
  4873. \end_layout
  4874. \end_inset
  4875. \end_layout
  4876. \begin_layout Subsection
  4877. Defined pattern analysis
  4878. \end_layout
  4879. \begin_layout Standard
  4880. \begin_inset Flex TODO Note (inline)
  4881. status open
  4882. \begin_layout Plain Layout
  4883. This was where I defined interesting expression patterns and then looked
  4884. at initial relative promoter coverage for each expression pattern.
  4885. Negative result.
  4886. I forgot about this until recently.
  4887. Worth including? Remember to also write methods.
  4888. \end_layout
  4889. \end_inset
  4890. \end_layout
  4891. \begin_layout Subsection
  4892. Promoter CpG islands?
  4893. \end_layout
  4894. \begin_layout Standard
  4895. \begin_inset Flex TODO Note (inline)
  4896. status collapsed
  4897. \begin_layout Plain Layout
  4898. I forgot until recently about the work I did on this.
  4899. Worth including? Remember to also write methods.
  4900. \end_layout
  4901. \end_inset
  4902. \end_layout
  4903. \begin_layout Section
  4904. Discussion
  4905. \end_layout
  4906. \begin_layout Standard
  4907. \begin_inset Flex TODO Note (inline)
  4908. status open
  4909. \begin_layout Plain Layout
  4910. Write better section headers
  4911. \end_layout
  4912. \end_inset
  4913. \end_layout
  4914. \begin_layout Subsection
  4915. Effective promoter radius
  4916. \end_layout
  4917. \begin_layout Standard
  4918. Figure
  4919. \begin_inset CommandInset ref
  4920. LatexCommand ref
  4921. reference "fig:near-promoter-peak-enrich"
  4922. plural "false"
  4923. caps "false"
  4924. noprefix "false"
  4925. \end_inset
  4926. shows that H3K4me2, H3K4me3, and H3K27me3 are all enriched near promoters,
  4927. relative to the rest of the genome, consistent with their conventionally
  4928. understood role in regulating gene transcription.
  4929. Interestingly, the radius within this enrichment occurs is not the same
  4930. for each histone mark.
  4931. H3K4me2 and H3K4me3 are enriched within a 1
  4932. \begin_inset space \thinspace{}
  4933. \end_inset
  4934. kb radius, while H3K27me3 is enriched within 2.5
  4935. \begin_inset space \thinspace{}
  4936. \end_inset
  4937. kb.
  4938. Notably, the determined promoter radius was consistent across all experimental
  4939. conditions, varying only between different histone marks.
  4940. This suggests that the conventional
  4941. \begin_inset Quotes eld
  4942. \end_inset
  4943. one size fits all
  4944. \begin_inset Quotes erd
  4945. \end_inset
  4946. approach of defining a single promoter region for each gene (or each TSS)
  4947. and using that same promoter region for analyzing all types of genomic
  4948. data within an experiment may not be appropriate, and a better approach
  4949. may be to use a separate promoter radius for each kind of data, with each
  4950. radius being derived from the data itself.
  4951. Furthermore, the apparent assymetry of upstream and downstream promoter
  4952. histone modification with respect to gene expression, seen in Figures
  4953. \begin_inset CommandInset ref
  4954. LatexCommand ref
  4955. reference "fig:H3K4me2-neighborhood"
  4956. plural "false"
  4957. caps "false"
  4958. noprefix "false"
  4959. \end_inset
  4960. ,
  4961. \begin_inset CommandInset ref
  4962. LatexCommand ref
  4963. reference "fig:H3K4me3-neighborhood"
  4964. plural "false"
  4965. caps "false"
  4966. noprefix "false"
  4967. \end_inset
  4968. , and
  4969. \begin_inset CommandInset ref
  4970. LatexCommand ref
  4971. reference "fig:H3K27me3-neighborhood"
  4972. plural "false"
  4973. caps "false"
  4974. noprefix "false"
  4975. \end_inset
  4976. , shows that even the concept of a promoter
  4977. \begin_inset Quotes eld
  4978. \end_inset
  4979. radius
  4980. \begin_inset Quotes erd
  4981. \end_inset
  4982. is likely an oversimplification.
  4983. At a minimum, nearby enrichment of peaks should be evaluated separately
  4984. for both upstream and downstream peaks, and an appropriate
  4985. \begin_inset Quotes eld
  4986. \end_inset
  4987. radius
  4988. \begin_inset Quotes erd
  4989. \end_inset
  4990. should be selected for each direction.
  4991. \end_layout
  4992. \begin_layout Standard
  4993. Figures
  4994. \begin_inset CommandInset ref
  4995. LatexCommand ref
  4996. reference "fig:H3K4me2-neighborhood"
  4997. plural "false"
  4998. caps "false"
  4999. noprefix "false"
  5000. \end_inset
  5001. and
  5002. \begin_inset CommandInset ref
  5003. LatexCommand ref
  5004. reference "fig:H3K4me3-neighborhood"
  5005. plural "false"
  5006. caps "false"
  5007. noprefix "false"
  5008. \end_inset
  5009. show that the determined promoter radius of 1
  5010. \begin_inset space ~
  5011. \end_inset
  5012. kb is approximately consistent with the distance from the TSS at which enrichmen
  5013. t of H3K4 methylationis correlates with increased expression, showing that
  5014. this radius, which was determined by a simple analysis of measuring the
  5015. distance from each TSS to the nearest peak, also has functional significance.
  5016. For H3K27me3, the correlation between histone modification near the promoter
  5017. and gene expression is more complex, involving non-peak variations such
  5018. as troughs in coverage at the TSS and asymmetric coverage upstream and
  5019. downstream, so it is difficult in this case to evaluate whether the 2.5
  5020. \begin_inset space ~
  5021. \end_inset
  5022. kb radius determined from TSS-to-peak distances is functionally significant.
  5023. However, the two patterns of coverage associated with elevated expression
  5024. levels both have interesting features within this radius.
  5025. \end_layout
  5026. \begin_layout Standard
  5027. \begin_inset Flex TODO Note (inline)
  5028. status open
  5029. \begin_layout Plain Layout
  5030. My instinct is to say
  5031. \begin_inset Quotes eld
  5032. \end_inset
  5033. further study is needed
  5034. \begin_inset Quotes erd
  5035. \end_inset
  5036. here, but that goes in Chapter 5, right?
  5037. \end_layout
  5038. \end_inset
  5039. \end_layout
  5040. \begin_layout Subsection
  5041. Convergence
  5042. \end_layout
  5043. \begin_layout Standard
  5044. \begin_inset Flex TODO Note (inline)
  5045. status open
  5046. \begin_layout Plain Layout
  5047. Look up some more references for these histone marks being involved in memory
  5048. differentiation.
  5049. (Ask Sarah)
  5050. \end_layout
  5051. \end_inset
  5052. \end_layout
  5053. \begin_layout Standard
  5054. We have observed that all 3 histone marks and the gene expression data all
  5055. exhibit evidence of convergence in abundance between naive and memory cells
  5056. by day 14 after activation (Figure
  5057. \begin_inset CommandInset ref
  5058. LatexCommand ref
  5059. reference "fig:PCoA-promoters"
  5060. plural "false"
  5061. caps "false"
  5062. noprefix "false"
  5063. \end_inset
  5064. , Table
  5065. \begin_inset CommandInset ref
  5066. LatexCommand ref
  5067. reference "tab:Number-signif-promoters"
  5068. plural "false"
  5069. caps "false"
  5070. noprefix "false"
  5071. \end_inset
  5072. ).
  5073. The MOFA latent factor scatter plots (Figure
  5074. \begin_inset CommandInset ref
  5075. LatexCommand ref
  5076. reference "fig:mofa-lf-scatter"
  5077. plural "false"
  5078. caps "false"
  5079. noprefix "false"
  5080. \end_inset
  5081. ) show that this pattern of convergence is captured in latent factor 5.
  5082. Like all the latent factors in this plot, this factor explains a substantial
  5083. portion of the variance in all 4 data sets, indicating a coordinated pattern
  5084. of variation shared across all histone marks and gene expression.
  5085. This, of course, is consistent with the expectation that any naive CD4
  5086. T-cells remaining at day 14 should have differentiated into memory cells
  5087. by that time, and should therefore have a genomic state similar to memory
  5088. cells.
  5089. This convergence is evidence that these histone marks all play an important
  5090. role in the naive-to-memory differentiation process.
  5091. A histone mark that was not involved in naive-to-memory differentiation
  5092. would not be expected to converge in this way after activation.
  5093. \end_layout
  5094. \begin_layout Standard
  5095. \begin_inset Float figure
  5096. wide false
  5097. sideways false
  5098. status collapsed
  5099. \begin_layout Plain Layout
  5100. \align center
  5101. \begin_inset Graphics
  5102. filename graphics/CD4-csaw/LaMere2016_fig8.pdf
  5103. lyxscale 50
  5104. width 60col%
  5105. groupId colwidth
  5106. \end_inset
  5107. \end_layout
  5108. \begin_layout Plain Layout
  5109. \begin_inset Caption Standard
  5110. \begin_layout Plain Layout
  5111. \series bold
  5112. \begin_inset CommandInset label
  5113. LatexCommand label
  5114. name "fig:Lamere2016-Fig8"
  5115. \end_inset
  5116. Lamere 2016 Figure 8
  5117. \begin_inset CommandInset citation
  5118. LatexCommand cite
  5119. key "LaMere2016"
  5120. literal "false"
  5121. \end_inset
  5122. ,
  5123. \begin_inset Quotes eld
  5124. \end_inset
  5125. Model for the role of H3K4 methylation during CD4 T-cell activation.
  5126. \begin_inset Quotes erd
  5127. \end_inset
  5128. \series default
  5129. Reproduced with permission.
  5130. \end_layout
  5131. \end_inset
  5132. \end_layout
  5133. \end_inset
  5134. \end_layout
  5135. \begin_layout Standard
  5136. In H3K4me2, H3K4me3, and RNA-seq, this convergence appears to be in progress
  5137. already by Day 5, shown by the smaller distance between naive and memory
  5138. cells at day 5 along the
  5139. \begin_inset Formula $y$
  5140. \end_inset
  5141. -axes in Figures
  5142. \begin_inset CommandInset ref
  5143. LatexCommand ref
  5144. reference "fig:PCoA-H3K4me2-prom"
  5145. plural "false"
  5146. caps "false"
  5147. noprefix "false"
  5148. \end_inset
  5149. ,
  5150. \begin_inset CommandInset ref
  5151. LatexCommand ref
  5152. reference "fig:PCoA-H3K4me3-prom"
  5153. plural "false"
  5154. caps "false"
  5155. noprefix "false"
  5156. \end_inset
  5157. , and
  5158. \begin_inset CommandInset ref
  5159. LatexCommand ref
  5160. reference "fig:RNA-PCA-group"
  5161. plural "false"
  5162. caps "false"
  5163. noprefix "false"
  5164. \end_inset
  5165. .
  5166. This agrees with the model proposed by Sarah Lamere based on an prior analysis
  5167. of the same data, shown in Figure
  5168. \begin_inset CommandInset ref
  5169. LatexCommand ref
  5170. reference "fig:Lamere2016-Fig8"
  5171. plural "false"
  5172. caps "false"
  5173. noprefix "false"
  5174. \end_inset
  5175. , which shows the pattern of H3K4 methylation and expression for naive cells
  5176. and memory cells converging at day 5.
  5177. This model was developed without the benefit of the PCoA plots in Figure
  5178. \begin_inset CommandInset ref
  5179. LatexCommand ref
  5180. reference "fig:PCoA-promoters"
  5181. plural "false"
  5182. caps "false"
  5183. noprefix "false"
  5184. \end_inset
  5185. , which have been corrected for confounding factors by ComBat and SVA.
  5186. This shows that proper batch correction assists in extracting meaningful
  5187. patterns in the data while eliminating systematic sources of irrelevant
  5188. variation in the data, allowing simple automated procedures like PCoA to
  5189. reveal interesting behaviors in the data that were previously only detectable
  5190. by a detailed manual analysis.
  5191. \end_layout
  5192. \begin_layout Standard
  5193. While the ideal comparison to demonstrate this convergence would be naive
  5194. cells at day 14 to memory cells at day 0, this is not feasible in this
  5195. experimental system, since neither naive nor memory cells are able to fully
  5196. return to their pre-activation state, as shown by the lack of overlap between
  5197. days 0 and 14 for either naive or memory cells in Figure
  5198. \begin_inset CommandInset ref
  5199. LatexCommand ref
  5200. reference "fig:PCoA-promoters"
  5201. plural "false"
  5202. caps "false"
  5203. noprefix "false"
  5204. \end_inset
  5205. .
  5206. \end_layout
  5207. \begin_layout Subsection
  5208. Positional
  5209. \end_layout
  5210. \begin_layout Standard
  5211. When looking at patterns in the relative coverage of each histone mark near
  5212. the TSS of each gene, several interesting patterns were apparent.
  5213. For H3K4me2 and H3K4me3, the pattern was straightforward: the consistent
  5214. pattern across all promoters was a single peak a few kb wide, with the
  5215. main axis of variation being the position of this peak relative to the
  5216. TSS (Figures
  5217. \begin_inset CommandInset ref
  5218. LatexCommand ref
  5219. reference "fig:H3K4me2-neighborhood"
  5220. plural "false"
  5221. caps "false"
  5222. noprefix "false"
  5223. \end_inset
  5224. &
  5225. \begin_inset CommandInset ref
  5226. LatexCommand ref
  5227. reference "fig:H3K4me3-neighborhood"
  5228. plural "false"
  5229. caps "false"
  5230. noprefix "false"
  5231. \end_inset
  5232. ).
  5233. There were no obvious
  5234. \begin_inset Quotes eld
  5235. \end_inset
  5236. preferred
  5237. \begin_inset Quotes erd
  5238. \end_inset
  5239. positions, but rather a continuous distribution of relative positions ranging
  5240. all across the promoter region.
  5241. The association with gene expression was also straightforward: peaks closer
  5242. to the TSS were more strongly associated with elevated gene expression.
  5243. Coverage downstream of the TSS appears to be more strongly associated with
  5244. elevated expression than coverage the same distance upstream, indicating
  5245. that the
  5246. \begin_inset Quotes eld
  5247. \end_inset
  5248. effective promoter region
  5249. \begin_inset Quotes erd
  5250. \end_inset
  5251. for H3K4me2 and H3K4me3 may be centered downstream of the TSS.
  5252. \end_layout
  5253. \begin_layout Standard
  5254. The relative promoter coverage for H3K27me3 had a more complex pattern,
  5255. with two specific patterns of promoter coverage associated with elevated
  5256. expression: a sharp depletion of H3K27me3 around the TSS relative to the
  5257. surrounding area, and a depletion of H3K27me3 downstream of the TSS relative
  5258. to upstream (Figure
  5259. \begin_inset CommandInset ref
  5260. LatexCommand ref
  5261. reference "fig:H3K27me3-neighborhood"
  5262. plural "false"
  5263. caps "false"
  5264. noprefix "false"
  5265. \end_inset
  5266. ).
  5267. A previous study found that H3K27me3 depletion within the gene body was
  5268. associated with elevated gene expression in 4 different cell types in mice
  5269. \begin_inset CommandInset citation
  5270. LatexCommand cite
  5271. key "Young2011"
  5272. literal "false"
  5273. \end_inset
  5274. .
  5275. This is consistent with the second pattern described here.
  5276. This study also reported that a spike in coverage at the TSS was associated
  5277. with
  5278. \emph on
  5279. lower
  5280. \emph default
  5281. expression, which is indirectly consistent with the first pattern described
  5282. here, in the sense that it associates lower H3K27me3 levels near the TSS
  5283. with higher expression.
  5284. \end_layout
  5285. \begin_layout Subsection
  5286. Workflow
  5287. \end_layout
  5288. \begin_layout Standard
  5289. \begin_inset ERT
  5290. status open
  5291. \begin_layout Plain Layout
  5292. \backslash
  5293. afterpage{
  5294. \end_layout
  5295. \begin_layout Plain Layout
  5296. \backslash
  5297. begin{landscape}
  5298. \end_layout
  5299. \end_inset
  5300. \end_layout
  5301. \begin_layout Standard
  5302. \begin_inset Float figure
  5303. wide false
  5304. sideways false
  5305. status open
  5306. \begin_layout Plain Layout
  5307. \align center
  5308. \begin_inset Graphics
  5309. filename graphics/CD4-csaw/rulegraphs/rulegraph-all.pdf
  5310. lyxscale 50
  5311. width 100col%
  5312. height 95theight%
  5313. \end_inset
  5314. \end_layout
  5315. \begin_layout Plain Layout
  5316. \begin_inset Caption Standard
  5317. \begin_layout Plain Layout
  5318. \begin_inset CommandInset label
  5319. LatexCommand label
  5320. name "fig:rulegraph"
  5321. \end_inset
  5322. \series bold
  5323. Dependency graph of steps in reproducible workflow.
  5324. \end_layout
  5325. \end_inset
  5326. \end_layout
  5327. \end_inset
  5328. \end_layout
  5329. \begin_layout Standard
  5330. \begin_inset ERT
  5331. status open
  5332. \begin_layout Plain Layout
  5333. \backslash
  5334. end{landscape}
  5335. \end_layout
  5336. \begin_layout Plain Layout
  5337. }
  5338. \end_layout
  5339. \end_inset
  5340. \end_layout
  5341. \begin_layout Standard
  5342. The analyses described in this chapter were organized into a reproducible
  5343. workflow using the Snakemake workflow management system.
  5344. As shown in Figure
  5345. \begin_inset CommandInset ref
  5346. LatexCommand ref
  5347. reference "fig:rulegraph"
  5348. plural "false"
  5349. caps "false"
  5350. noprefix "false"
  5351. \end_inset
  5352. , the workflow includes many steps with complex dependencies between them.
  5353. For example, the step that counts the number of ChIP-seq reads in 500
  5354. \begin_inset space ~
  5355. \end_inset
  5356. bp windows in each promoter (the starting point for Figures
  5357. \begin_inset CommandInset ref
  5358. LatexCommand ref
  5359. reference "fig:H3K4me2-neighborhood"
  5360. plural "false"
  5361. caps "false"
  5362. noprefix "false"
  5363. \end_inset
  5364. ,
  5365. \begin_inset CommandInset ref
  5366. LatexCommand ref
  5367. reference "fig:H3K4me3-neighborhood"
  5368. plural "false"
  5369. caps "false"
  5370. noprefix "false"
  5371. \end_inset
  5372. , and
  5373. \begin_inset CommandInset ref
  5374. LatexCommand ref
  5375. reference "fig:H3K27me3-neighborhood"
  5376. plural "false"
  5377. caps "false"
  5378. noprefix "false"
  5379. \end_inset
  5380. ), named
  5381. \begin_inset Formula $\texttt{chipseq\_count\_tss\_neighborhoods}$
  5382. \end_inset
  5383. , depends on the RNA-seq abundance estimates in order to select the most-used
  5384. TSS for each gene, the aligned ChIP-seq reads, the index for those reads,
  5385. and the blacklist of regions to be excluded from ChIP-seq analysis.
  5386. Each step declares its inputs and outputs, and Snakemake uses these to
  5387. determine the dependencies between steps.
  5388. Each step is marked as depending on all the steps whose outputs match its
  5389. inputs, generating the workflow graph in Figure
  5390. \begin_inset CommandInset ref
  5391. LatexCommand ref
  5392. reference "fig:rulegraph"
  5393. plural "false"
  5394. caps "false"
  5395. noprefix "false"
  5396. \end_inset
  5397. , which Snakemake uses to determine order in which to execute each step
  5398. so that each step is executed only after all of the steps it depends on
  5399. have completed, thereby automating the entire workflow from start to finish.
  5400. \end_layout
  5401. \begin_layout Standard
  5402. In addition to simply making it easier to organize the steps in the analysis,
  5403. structuring the analysis as a workflow allowed for some analysis strategies
  5404. that would not have been practical otherwise.
  5405. For example, 5 different RNA-seq quantification methods were tested against
  5406. two different reference transcriptome annotations for a total of 10 different
  5407. quantifications of the same RNA-seq data.
  5408. These were then compared against each other in the exploratory data analysis
  5409. step, to determine that the results were not very sensitive to either the
  5410. choice of quantification method or the choice of annotation.
  5411. This was possible with a single script for the exploratory data analysis,
  5412. because Snakemake was able to automate running this script for every combinatio
  5413. n of method and reference.
  5414. In a similar manner, two different peak calling methods were tested against
  5415. each other, and in this case it was determined that SICER was unambiguously
  5416. superior to MACS for all histone marks studied.
  5417. By enabling these types of comparisons, structuring the analysis as an
  5418. automated workflow allowed important analysis decisions to be made in a
  5419. data-driven way, by running every reasonable option through the downstream
  5420. steps, seeing the consequences of choosing each option, and deciding accordingl
  5421. y.
  5422. \end_layout
  5423. \begin_layout Subsection
  5424. Data quality issues limit conclusions
  5425. \end_layout
  5426. \begin_layout Standard
  5427. \begin_inset Flex TODO Note (inline)
  5428. status open
  5429. \begin_layout Plain Layout
  5430. Is this needed?
  5431. \end_layout
  5432. \end_inset
  5433. \end_layout
  5434. \begin_layout Section
  5435. Future Directions
  5436. \end_layout
  5437. \begin_layout Standard
  5438. The analysis of RNA-seq and ChIP-seq in CD4 T-cells in Chapter 2 is in many
  5439. ways a preliminary study that suggests a multitude of new avenues of investigat
  5440. ion.
  5441. Here we consider a selection of such avenues.
  5442. \end_layout
  5443. \begin_layout Subsection
  5444. Negative results
  5445. \end_layout
  5446. \begin_layout Standard
  5447. Two additional analyses were conducted beyond those reported in the results.
  5448. First, we searched for evidence that the presence or absence of a CpG island
  5449. in the promoter was correlated with increases or decreases in gene expression
  5450. or any histone mark in any of the tested contrasts.
  5451. Second, we searched for evidence that the relative ChIP-seq coverage profiles
  5452. prior to activations could predict the change in expression of a gene after
  5453. activation.
  5454. Neither analysis turned up any clear positive results.
  5455. \end_layout
  5456. \begin_layout Subsection
  5457. Improve on the idea of an effective promoter radius
  5458. \end_layout
  5459. \begin_layout Standard
  5460. This study introduced the concept of an
  5461. \begin_inset Quotes eld
  5462. \end_inset
  5463. effective promoter radius
  5464. \begin_inset Quotes erd
  5465. \end_inset
  5466. specific to each histone mark based on distince from the TSS within which
  5467. an excess of peaks was called for that mark.
  5468. This concept was then used to guide further analyses throughout the study.
  5469. However, while the effective promoter radius was useful in those analyses,
  5470. it is both limited in theory and shown in practice to be a possible oversimplif
  5471. ication.
  5472. First, the effective promoter radii used in this study were chosen based
  5473. on manual inspection of the TSS-to-peak distance distributions in Figure
  5474. \begin_inset CommandInset ref
  5475. LatexCommand ref
  5476. reference "fig:near-promoter-peak-enrich"
  5477. plural "false"
  5478. caps "false"
  5479. noprefix "false"
  5480. \end_inset
  5481. , selecting round numbers of analyst convenience (Table
  5482. \begin_inset CommandInset ref
  5483. LatexCommand ref
  5484. reference "tab:effective-promoter-radius"
  5485. plural "false"
  5486. caps "false"
  5487. noprefix "false"
  5488. \end_inset
  5489. ).
  5490. It would be better to define an algorithm that selects a more precise radius
  5491. based on the features of the graph.
  5492. One possible way to do this would be to randomly rearrange the called peaks
  5493. throughout the genome many (while preserving the distribution of peak widths)
  5494. and re-generate the same plot as in Figure
  5495. \begin_inset CommandInset ref
  5496. LatexCommand ref
  5497. reference "fig:near-promoter-peak-enrich"
  5498. plural "false"
  5499. caps "false"
  5500. noprefix "false"
  5501. \end_inset
  5502. .
  5503. This would yield a better
  5504. \begin_inset Quotes eld
  5505. \end_inset
  5506. background
  5507. \begin_inset Quotes erd
  5508. \end_inset
  5509. distribution that demonstrates the degree of near-TSS enrichment that would
  5510. be expected by random chance.
  5511. The effective promoter radius could be defined as the point where the true
  5512. distribution diverges from the randomized background distribution.
  5513. \end_layout
  5514. \begin_layout Standard
  5515. Furthermore, the above definition of effective promoter radius has the significa
  5516. nt limitation of being based on the peak calling method.
  5517. It is thus very sensitive to the choice of peak caller and significance
  5518. threshold for calling peaks, as well as the degree of saturation in the
  5519. sequencing.
  5520. Calling peaks from ChIP-seq samples with insufficient coverage depth, with
  5521. the wrong peak caller, or with a different significance threshold could
  5522. give a drastically different number of called peaks, and hence a drastically
  5523. different distribution of peak-to-TSS distances.
  5524. To address this, it is desirable to develop a better method of determining
  5525. the effective promoter radius that relies only on the distribution of read
  5526. coverage around the TSS, independent of the peak calling.
  5527. Furthermore, as demonstrated by the upstream-downstream asymmetries observed
  5528. in Figures
  5529. \begin_inset CommandInset ref
  5530. LatexCommand ref
  5531. reference "fig:H3K4me2-neighborhood"
  5532. plural "false"
  5533. caps "false"
  5534. noprefix "false"
  5535. \end_inset
  5536. ,
  5537. \begin_inset CommandInset ref
  5538. LatexCommand ref
  5539. reference "fig:H3K4me3-neighborhood"
  5540. plural "false"
  5541. caps "false"
  5542. noprefix "false"
  5543. \end_inset
  5544. , and
  5545. \begin_inset CommandInset ref
  5546. LatexCommand ref
  5547. reference "fig:H3K27me3-neighborhood"
  5548. plural "false"
  5549. caps "false"
  5550. noprefix "false"
  5551. \end_inset
  5552. , this definition should determine a different radius for the upstream and
  5553. downstream directions.
  5554. At this point, it may be better to rename this concept
  5555. \begin_inset Quotes eld
  5556. \end_inset
  5557. effective promoter extent
  5558. \begin_inset Quotes erd
  5559. \end_inset
  5560. and avoid the word
  5561. \begin_inset Quotes eld
  5562. \end_inset
  5563. radius
  5564. \begin_inset Quotes erd
  5565. \end_inset
  5566. , since a radius implies a symmetry about the TSS that is not supported
  5567. by the data.
  5568. \end_layout
  5569. \begin_layout Standard
  5570. Beyond improving the definition of effective promoter extent, functional
  5571. validation is necessary to show that this measure of near-TSS enrichment
  5572. has biological meaning.
  5573. Figures
  5574. \begin_inset CommandInset ref
  5575. LatexCommand ref
  5576. reference "fig:H3K4me2-neighborhood"
  5577. plural "false"
  5578. caps "false"
  5579. noprefix "false"
  5580. \end_inset
  5581. and
  5582. \begin_inset CommandInset ref
  5583. LatexCommand ref
  5584. reference "fig:H3K4me3-neighborhood"
  5585. plural "false"
  5586. caps "false"
  5587. noprefix "false"
  5588. \end_inset
  5589. already provide a very limited functional validation of the chosen promoter
  5590. extents for H3K4me2 and H3K4me3 by showing that spikes in coverage within
  5591. this region are most strongly correlated with elevated gene expression.
  5592. However, there are other ways to show functional relevance of the promoter
  5593. extent.
  5594. For example, correlations could be computed between read counts in peaks
  5595. nearby gene promoters and the expression level of those genes, and these
  5596. correlations could be plotted against the distance of the peak upstream
  5597. or downstream of the gene's TSS.
  5598. If the promoter extent truly defines a
  5599. \begin_inset Quotes eld
  5600. \end_inset
  5601. sphere of influence
  5602. \begin_inset Quotes erd
  5603. \end_inset
  5604. within which a histone mark is involved with the regulation of a gene,
  5605. then the correlations for peaks within this extent should be significantly
  5606. higher than those further upstream or downstream.
  5607. Peaks within these extents may also be more likely to show differential
  5608. modification than those outside genic regions of the genome.
  5609. \end_layout
  5610. \begin_layout Subsection
  5611. Design experiments to focus on post-activation convergence of naive & memory
  5612. cells
  5613. \end_layout
  5614. \begin_layout Standard
  5615. In this study, a convergence between naive and memory cells was observed
  5616. in both the pattern of gene expression and in epigenetic state of the 3
  5617. histone marks studied, consistent with the hypothesis that any naive cells
  5618. remaining 14 days after activation have differentiated into memory cells,
  5619. and that both gene expression and these histone marks are involved in this
  5620. differentiation.
  5621. However, the current study was not designed with this specific hypothesis
  5622. in mind, and it therefore has some deficiencies with regard to testing
  5623. it.
  5624. The memory CD4 samples at day 14 do not resemble the memory samples at
  5625. day 0, indicating that in the specific model of activation used for this
  5626. experiment, the cells are not guaranteed to return to their original pre-activa
  5627. tion state, or perhaps this process takes substantially longer than 14 days.
  5628. This is a challenge for the convergence hypothesis because the ideal comparison
  5629. to prove that naive cells are converging to a resting memory state would
  5630. be to compare the final naive time point to the Day 0 memory samples, but
  5631. this comparison is only meaningful if memory cells generally return to
  5632. the same
  5633. \begin_inset Quotes eld
  5634. \end_inset
  5635. resting
  5636. \begin_inset Quotes erd
  5637. \end_inset
  5638. state that they started at.
  5639. \end_layout
  5640. \begin_layout Standard
  5641. To better study the convergence hypothesis, a new experiment should be designed
  5642. using a model system for T-cell activation that is known to allow cells
  5643. to return as closely as possible to their pre-activation state.
  5644. Alternatively, if it is not possible to find or design such a model system,
  5645. the same cell cultures could be activated serially multiple times, and
  5646. sequenced after each activation cycle right before the next activation.
  5647. It is likely that several activations in the same model system will settle
  5648. into a cylical pattern, converging to a consistent
  5649. \begin_inset Quotes eld
  5650. \end_inset
  5651. resting
  5652. \begin_inset Quotes erd
  5653. \end_inset
  5654. state after each activation, even if this state is different from the initial
  5655. resting state at Day 0.
  5656. If so, it will be possible to compare the final states of both naive and
  5657. memory cells to show that they converge despite different initial conditions.
  5658. \end_layout
  5659. \begin_layout Standard
  5660. In addition, if naive-to-memory convergence is a general pattern, it should
  5661. also be detectable in other epigenetic marks, including other histone marks
  5662. and DNA methylation.
  5663. An experiment should be designed studying a large number of epigenetic
  5664. marks known or suspected to be involved in regulation of gene expression,
  5665. assaying all of these at the same pre- and post-activation time points.
  5666. Multi-dataset factor analysis methods like MOFA can then be used to identify
  5667. coordinated patterns of regulation shared across many epigenetic marks.
  5668. If possible, some
  5669. \begin_inset Quotes eld
  5670. \end_inset
  5671. negative control
  5672. \begin_inset Quotes erd
  5673. \end_inset
  5674. marks should be included that are known
  5675. \emph on
  5676. not
  5677. \emph default
  5678. to be involved in T-cell activation or memory formation.
  5679. Of course, CD4 T-cells are not the only adaptive immune cells with memory.
  5680. A similar study could be designed for CD8 T-cells, B-cells, and even specific
  5681. subsets of CD4 T-cells.
  5682. \end_layout
  5683. \begin_layout Subsection
  5684. Follow up on hints of interesting patterns in promoter relative coverage
  5685. profiles
  5686. \end_layout
  5687. \begin_layout Standard
  5688. \begin_inset Flex TODO Note (inline)
  5689. status open
  5690. \begin_layout Plain Layout
  5691. I think I might need to write up the negative results for the Promoter CpG
  5692. and defined pattern analysis before writing this section.
  5693. \end_layout
  5694. \end_inset
  5695. \end_layout
  5696. \begin_layout Itemize
  5697. Also find better normalizations: maybe borrow from MACS/SICER background
  5698. correction methods?
  5699. \end_layout
  5700. \begin_layout Itemize
  5701. For H3K4, define polar coordinates based on PC1 & 2: R = peak size, Theta
  5702. = peak position.
  5703. Then correlate with expression.
  5704. \end_layout
  5705. \begin_layout Itemize
  5706. Current analysis only at Day 0.
  5707. Need to study across time points.
  5708. \end_layout
  5709. \begin_layout Itemize
  5710. Integrating data across so many dimensions is a significant analysis challenge
  5711. \end_layout
  5712. \begin_layout Subsection
  5713. Investigate causes of high correlation between mutually exclusive histone
  5714. marks
  5715. \end_layout
  5716. \begin_layout Standard
  5717. The high correlation between coverage depth observed between H3K4me2 and
  5718. H3K4me3 is both expected and unexpected.
  5719. Since both marks are associated with elevated gene transcription, a positive
  5720. correlation between them is not surprising.
  5721. However, these two marks represent different post-translational modifications
  5722. of the
  5723. \emph on
  5724. same
  5725. \emph default
  5726. lysine residue on the histone H3 polypeptide, which means that they cannot
  5727. both be present on the same H3 subunit.
  5728. Thus, the high correlation between them has several potential explanations.
  5729. One possible reason is cell population heterogeneity: perhaps some genomic
  5730. loci are frequently marked with H3K4me2 in some cells, while in other cells
  5731. the same loci are marked with H3K4me3.
  5732. Another possibility is allele-specific modifications: the loci are marked
  5733. in each diploid cell with H3K4me2 on one allele and H3K4me3 on the other
  5734. allele.
  5735. Lastly, since each histone octamer contains 2 H3 subunits, it is possible
  5736. that having one H3K4me2 mark and one H3K4me3 mark on a given histone octamer
  5737. represents a distinct epigenetic state with a different function than either
  5738. double H3K4me2 or double H3K4me3.
  5739. \end_layout
  5740. \begin_layout Standard
  5741. These three hypotheses could be disentangled by single-cell ChIP-seq.
  5742. If the correlation between these two histone marks persists even within
  5743. the reads for each individual cell, then cell population heterogeneity
  5744. cannot explain the correlation.
  5745. Allele-specific modification can be tested for by looking at the correlation
  5746. between read coverage of the two histone marks at heterozygous loci.
  5747. If the correlation between read counts for opposite loci is low, then this
  5748. is consistent with allele-specific modification.
  5749. Finally if the modifications do not separate by either cell or allele,
  5750. the colocation of these two marks is most likely occurring at the level
  5751. of individual histones, with the heterogenously modified histone representing
  5752. a distinct state.
  5753. \end_layout
  5754. \begin_layout Standard
  5755. However, another experiment would be required to show direct evidence of
  5756. such a heterogeneously modified state.
  5757. Specifically a
  5758. \begin_inset Quotes eld
  5759. \end_inset
  5760. double ChIP
  5761. \begin_inset Quotes erd
  5762. \end_inset
  5763. experiment would need to be performed, where the input DNA is first subjected
  5764. to an immunoprecipitation pulldown from the anti-H3K4me2 antibody, and
  5765. then the enriched material is collected, with proteins still bound, and
  5766. immunoprecipitated
  5767. \emph on
  5768. again
  5769. \emph default
  5770. using the anti-H3K4me3 antibody.
  5771. If this yields significant numbers of non-artifactual reads in the same
  5772. regions as the individual pulldowns of the two marks, this is strong evidence
  5773. that the two marks are occurring on opposite H3 subunits of the same histones.
  5774. \end_layout
  5775. \begin_layout Standard
  5776. \begin_inset Flex TODO Note (inline)
  5777. status open
  5778. \begin_layout Plain Layout
  5779. Try to see if double ChIP-seq is actually feasible, and if not, come up
  5780. with some other idea for directly detecting the mixed mod state.
  5781. Oh! Actually ChIP-seq isn't required, only double ChIP followed by quantificati
  5782. on.
  5783. That's one possible angle.
  5784. \end_layout
  5785. \end_inset
  5786. \end_layout
  5787. \begin_layout Chapter
  5788. Improving array-based diagnostics for transplant rejection by optimizing
  5789. data preprocessing
  5790. \end_layout
  5791. \begin_layout Standard
  5792. \begin_inset Note Note
  5793. status open
  5794. \begin_layout Plain Layout
  5795. Chapter author list: Me, Sunil, Tom, Padma, Dan
  5796. \end_layout
  5797. \end_inset
  5798. \end_layout
  5799. \begin_layout Section
  5800. Approach
  5801. \end_layout
  5802. \begin_layout Subsection
  5803. Proper pre-processing is essential for array data
  5804. \end_layout
  5805. \begin_layout Standard
  5806. \begin_inset Flex TODO Note (inline)
  5807. status open
  5808. \begin_layout Plain Layout
  5809. This section could probably use some citations
  5810. \end_layout
  5811. \end_inset
  5812. \end_layout
  5813. \begin_layout Standard
  5814. Microarrays, bead arrays, and similar assays produce raw data in the form
  5815. of fluorescence intensity measurements, with the each intensity measurement
  5816. proportional to the abundance of some fluorescently-labelled target DNA
  5817. or RNA sequence that base pairs to a specific probe sequence.
  5818. However, these measurements for each probe are also affected my many technical
  5819. confounding factors, such as the concentration of target material, strength
  5820. of off-target binding, and the sensitivity of the imaging sensor.
  5821. Some array designs also use multiple probe sequences for each target.
  5822. Hence, extensive pre-processing of array data is necessary to normalize
  5823. out the effects of these technical factors and summarize the information
  5824. from multiple probes to arrive at a single usable estimate of abundance
  5825. or other relevant quantity, such as a ratio of two abundances, for each
  5826. target.
  5827. \end_layout
  5828. \begin_layout Standard
  5829. The choice of pre-processing algorithms used in the analysis of an array
  5830. data set can have a large effect on the results of that analysis.
  5831. However, despite their importance, these steps are often neglected or rushed
  5832. in order to get to the more scientifically interesting analysis steps involving
  5833. the actual biology of the system under study.
  5834. Hence, it is often possible to achieve substantial gains in statistical
  5835. power, model goodness-of-fit, or other relevant performance measures, by
  5836. checking the assumptions made by each preprocessing step and choosing specific
  5837. normalization methods tailored to the specific goals of the current analysis.
  5838. \end_layout
  5839. \begin_layout Subsection
  5840. Clinical diagnostic applications for microarrays require single-channel
  5841. normalization
  5842. \end_layout
  5843. \begin_layout Standard
  5844. As the cost of performing microarray assays falls, there is increasing interest
  5845. in using genomic assays for diagnostic purposes, such as distinguishing
  5846. healthy transplants (TX) from transplants undergoing acute rejection (AR)
  5847. or acute dysfunction with no rejection (ADNR).
  5848. However, the the standard normalization algorithm used for microarray data,
  5849. Robust Multi-chip Average (RMA)
  5850. \begin_inset CommandInset citation
  5851. LatexCommand cite
  5852. key "Irizarry2003a"
  5853. literal "false"
  5854. \end_inset
  5855. , is not applicable in a clinical setting.
  5856. Two of the steps in RMA, quantile normalization and probe summarization
  5857. by median polish, depend on every array in the data set being normalized.
  5858. This means that adding or removing any arrays from a data set changes the
  5859. normalized values for all arrays, and data sets that have been normalized
  5860. separately cannot be compared to each other.
  5861. Hence, when using RMA, any arrays to be analyzed together must also be
  5862. normalized together, and the set of arrays included in the data set must
  5863. be held constant throughout an analysis.
  5864. \end_layout
  5865. \begin_layout Standard
  5866. These limitations present serious impediments to the use of arrays as a
  5867. diagnostic tool.
  5868. When training a classifier, the samples to be classified must not be involved
  5869. in any step of the training process, lest their inclusion bias the training
  5870. process.
  5871. Once a classifier is deployed in a clinical setting, the samples to be
  5872. classified will not even
  5873. \emph on
  5874. exist
  5875. \emph default
  5876. at the time of training, so including them would be impossible even if
  5877. it were statistically justifiable.
  5878. Therefore, any machine learning application for microarrays demands that
  5879. the normalized expression values computed for an array must depend only
  5880. on information contained within that array.
  5881. This would ensure that each array's normalization is independent of every
  5882. other array, and that arrays normalized separately can still be compared
  5883. to each other without bias.
  5884. Such a normalization is commonly referred to as
  5885. \begin_inset Quotes eld
  5886. \end_inset
  5887. single-channel normalization
  5888. \begin_inset Quotes erd
  5889. \end_inset
  5890. .
  5891. \end_layout
  5892. \begin_layout Standard
  5893. Frozen RMA (fRMA) addresses these concerns by replacing the quantile normalizati
  5894. on and median polish with alternatives that do not introduce inter-array
  5895. dependence, allowing each array to be normalized independently of all others
  5896. \begin_inset CommandInset citation
  5897. LatexCommand cite
  5898. key "McCall2010"
  5899. literal "false"
  5900. \end_inset
  5901. .
  5902. Quantile normalization is performed against a pre-generated set of quantiles
  5903. learned from a collection of 850 publically available arrays sampled from
  5904. a wide variety of tissues in the Gene Expression Omnibus (GEO).
  5905. Each array's probe intensity distribution is normalized against these pre-gener
  5906. ated quantiles.
  5907. The median polish step is replaced with a robust weighted average of probe
  5908. intensities, using inverse variance weights learned from the same public
  5909. GEO data.
  5910. The result is a normalization that satisfies the requirements mentioned
  5911. above: each array is normalized independently of all others, and any two
  5912. normalized arrays can be compared directly to each other.
  5913. \end_layout
  5914. \begin_layout Standard
  5915. One important limitation of fRMA is that it requires a separate reference
  5916. data set from which to learn the parameters (reference quantiles and probe
  5917. weights) that will be used to normalize each array.
  5918. These parameters are specific to a given array platform, and pre-generated
  5919. parameters are only provided for the most common platforms, such as Affymetrix
  5920. hgu133plus2.
  5921. For a less common platform, such as hthgu133pluspm, is is necessary to
  5922. learn custom parameters from in-house data before fRMA can be used to normalize
  5923. samples on that platform
  5924. \begin_inset CommandInset citation
  5925. LatexCommand cite
  5926. key "McCall2011"
  5927. literal "false"
  5928. \end_inset
  5929. .
  5930. \end_layout
  5931. \begin_layout Standard
  5932. One other option is the aptly-named Single Channel Array Normalization (SCAN),
  5933. which adapts a normalization method originally designed for tiling arrays
  5934. \begin_inset CommandInset citation
  5935. LatexCommand cite
  5936. key "Piccolo2012"
  5937. literal "false"
  5938. \end_inset
  5939. .
  5940. SCAN is truly single-channel in that it does not require a set of normalization
  5941. paramters estimated from an external set of reference samples like fRMA
  5942. does.
  5943. \end_layout
  5944. \begin_layout Subsection
  5945. Heteroskedasticity must be accounted for in methylation array data
  5946. \end_layout
  5947. \begin_layout Standard
  5948. DNA methylation arrays are a relatively new kind of assay that uses microarrays
  5949. to measure the degree of methylation on cytosines in specific regions arrayed
  5950. across the genome.
  5951. First, bisulfite treatment converts all unmethylated cytosines to uracil
  5952. (which then become thymine after amplication) while leaving methylated
  5953. cytosines unaffected.
  5954. Then, each target region is interrogated with two probes: one binds to
  5955. the original genomic sequence and interrogates the level of methylated
  5956. DNA, and the other binds to the same sequence with all cytosines replaced
  5957. by thymidines and interrogates the level of unmethylated DNA.
  5958. \end_layout
  5959. \begin_layout Standard
  5960. \begin_inset Float figure
  5961. wide false
  5962. sideways false
  5963. status collapsed
  5964. \begin_layout Plain Layout
  5965. \align center
  5966. \begin_inset Graphics
  5967. filename graphics/methylvoom/sigmoid.pdf
  5968. lyxscale 50
  5969. width 60col%
  5970. groupId colwidth
  5971. \end_inset
  5972. \end_layout
  5973. \begin_layout Plain Layout
  5974. \begin_inset Caption Standard
  5975. \begin_layout Plain Layout
  5976. \begin_inset CommandInset label
  5977. LatexCommand label
  5978. name "fig:Sigmoid-beta-m-mapping"
  5979. \end_inset
  5980. \series bold
  5981. Sigmoid shape of the mapping between β and M values
  5982. \end_layout
  5983. \end_inset
  5984. \end_layout
  5985. \end_inset
  5986. \end_layout
  5987. \begin_layout Standard
  5988. After normalization, these two probe intensities are summarized in one of
  5989. two ways, each with advantages and disadvantages.
  5990. β
  5991. \series bold
  5992. \series default
  5993. values, interpreted as fraction of DNA copies methylated, range from 0 to
  5994. 1.
  5995. β
  5996. \series bold
  5997. \series default
  5998. values are conceptually easy to interpret, but the constrained range makes
  5999. them unsuitable for linear modeling, and their error distributions are
  6000. highly non-normal, which also frustrates linear modeling.
  6001. M-values, interpreted as the log ratio of methylated to unmethylated copies,
  6002. are computed by mapping the beta values from
  6003. \begin_inset Formula $[0,1]$
  6004. \end_inset
  6005. onto
  6006. \begin_inset Formula $(-\infty,+\infty)$
  6007. \end_inset
  6008. using a sigmoid curve (Figure
  6009. \begin_inset CommandInset ref
  6010. LatexCommand ref
  6011. reference "fig:Sigmoid-beta-m-mapping"
  6012. plural "false"
  6013. caps "false"
  6014. noprefix "false"
  6015. \end_inset
  6016. ).
  6017. This transformation results in values with better statistical perperties:
  6018. the unconstrained range is suitable for linear modeling, and the error
  6019. distributions are more normal.
  6020. Hence, most linear modeling and other statistical testing on methylation
  6021. arrays is performed using M-values.
  6022. \end_layout
  6023. \begin_layout Standard
  6024. However, the steep slope of the sigmoid transformation near 0 and 1 tends
  6025. to over-exaggerate small differences in β values near those extremes, which
  6026. in turn amplifies the error in those values, leading to a U-shaped trend
  6027. in the mean-variance curve: extreme values have higher variances than values
  6028. near the middle.
  6029. This mean-variance dependency must be accounted for when fitting the linear
  6030. model for differential methylation, or else the variance will be systematically
  6031. overestimated for probes with moderate M-values and underestimated for
  6032. probes with extreme M-values.
  6033. This is particularly undesirable for methylation data because the intermediate
  6034. M-values are the ones of most interest, since they are more likely to represent
  6035. areas of varying methylation, whereas extreme M-values typically represent
  6036. complete methylation or complete lack of methylation.
  6037. \end_layout
  6038. \begin_layout Standard
  6039. RNA-seq read count data are also known to show heteroskedasticity, and the
  6040. voom method was introduced for modeling this heteroskedasticity by estimating
  6041. the mean-variance trend in the data and using this trend to assign precision
  6042. weights to each observation
  6043. \begin_inset CommandInset citation
  6044. LatexCommand cite
  6045. key "Law2013"
  6046. literal "false"
  6047. \end_inset
  6048. .
  6049. While methylation array data are not derived from counts and have a very
  6050. different mean-variance relationship from that of typical RNA-seq data,
  6051. the voom method makes no specific assumptions on the shape of the mean-variance
  6052. relationship – it only assumes that the relationship can be modeled as
  6053. a smooth curve.
  6054. Hence, the method is sufficiently general to model the mean-variance relationsh
  6055. ip in methylation array data.
  6056. However, the standard implementation of voom assumes that the input is
  6057. given in raw read counts, and it must be adapted to run on methylation
  6058. M-values.
  6059. \end_layout
  6060. \begin_layout Section
  6061. Methods
  6062. \end_layout
  6063. \begin_layout Subsection
  6064. Evaluation of classifier performance with different normalization methods
  6065. \end_layout
  6066. \begin_layout Standard
  6067. For testing different expression microarray normalizations, a data set of
  6068. 157 hgu133plus2 arrays was used, consisting of blood samples from kidney
  6069. transplant patients whose grafts had been graded as TX, AR, or ADNR via
  6070. biopsy and histology (46 TX, 69 AR, 42 ADNR)
  6071. \begin_inset CommandInset citation
  6072. LatexCommand cite
  6073. key "Kurian2014"
  6074. literal "true"
  6075. \end_inset
  6076. .
  6077. Additionally, an external validation set of 75 samples was gathered from
  6078. public GEO data (37 TX, 38 AR, no ADNR).
  6079. \end_layout
  6080. \begin_layout Standard
  6081. \begin_inset Flex TODO Note (inline)
  6082. status open
  6083. \begin_layout Plain Layout
  6084. Find appropriate GEO identifiers if possible.
  6085. Kurian 2014 says GSE15296, but this seems to be different data.
  6086. I also need to look up the GEO accession for the external validation set.
  6087. \end_layout
  6088. \end_inset
  6089. \end_layout
  6090. \begin_layout Standard
  6091. To evaluate the effect of each normalization on classifier performance,
  6092. the same classifier training and validation procedure was used after each
  6093. normalization method.
  6094. The PAM package was used to train a nearest shrunken centroid classifier
  6095. on the training set and select the appropriate threshold for centroid shrinking.
  6096. Then the trained classifier was used to predict the class probabilities
  6097. of each validation sample.
  6098. From these class probabilities, ROC curves and area-under-curve (AUC) values
  6099. were generated
  6100. \begin_inset CommandInset citation
  6101. LatexCommand cite
  6102. key "Turck2011"
  6103. literal "false"
  6104. \end_inset
  6105. .
  6106. Each normalization was tested on two different sets of training and validation
  6107. samples.
  6108. For internal validation, the 115 TX and AR arrays in the internal set were
  6109. split at random into two equal sized sets, one for training and one for
  6110. validation, each containing the same numbers of TX and AR samples as the
  6111. other set.
  6112. For external validation, the full set of 115 TX and AR samples were used
  6113. as a training set, and the 75 external TX and AR samples were used as the
  6114. validation set.
  6115. Thus, 2 ROC curves and AUC values were generated for each normalization
  6116. method: one internal and one external.
  6117. Because the external validation set contains no ADNR samples, only classificati
  6118. on of TX and AR samples was considered.
  6119. The ADNR samples were included during normalization but excluded from all
  6120. classifier training and validation.
  6121. This ensures that the performance on internal and external validation sets
  6122. is directly comparable, since both are performing the same task: distinguising
  6123. TX from AR.
  6124. \end_layout
  6125. \begin_layout Standard
  6126. \begin_inset Flex TODO Note (inline)
  6127. status open
  6128. \begin_layout Plain Layout
  6129. Summarize the get.best.threshold algorithm for PAM threshold selection, or
  6130. just put the code online?
  6131. \end_layout
  6132. \end_inset
  6133. \end_layout
  6134. \begin_layout Standard
  6135. Six different normalization strategies were evaluated.
  6136. First, 2 well-known non-single-channel normalization methods were considered:
  6137. RMA and dChip
  6138. \begin_inset CommandInset citation
  6139. LatexCommand cite
  6140. key "Li2001,Irizarry2003a"
  6141. literal "false"
  6142. \end_inset
  6143. .
  6144. Since RMA produces expression values on a log2 scale and dChip does not,
  6145. the values from dChip were log2 transformed after normalization.
  6146. Next, RMA and dChip followed by Global Rank-invariant Set Normalization
  6147. (GRSN) were tested
  6148. \begin_inset CommandInset citation
  6149. LatexCommand cite
  6150. key "Pelz2008"
  6151. literal "false"
  6152. \end_inset
  6153. .
  6154. Post-processing with GRSN does not turn RMA or dChip into single-channel
  6155. methods, but it may help mitigate batch effects and is therefore useful
  6156. as a benchmark.
  6157. Lastly, the two single-channel normalization methods, fRMA and SCAN, were
  6158. tested
  6159. \begin_inset CommandInset citation
  6160. LatexCommand cite
  6161. key "McCall2010,Piccolo2012"
  6162. literal "false"
  6163. \end_inset
  6164. .
  6165. When evaluting internal validation performance, only the 157 internal samples
  6166. were normalized; when evaluating external validation performance, all 157
  6167. internal samples and 75 external samples were normalized together.
  6168. \end_layout
  6169. \begin_layout Standard
  6170. For demonstrating the problem with separate normalization of training and
  6171. validation data, one additional normalization was performed: the internal
  6172. and external sets were each normalized separately using RMA, and the normalized
  6173. data for each set were combined into a single set with no further attempts
  6174. at normalizing between the two sets.
  6175. The represents approximately how RMA would have to be used in a clinical
  6176. setting, where the samples to be classified are not available at the time
  6177. the classifier is trained.
  6178. \end_layout
  6179. \begin_layout Subsection
  6180. Generating custom fRMA vectors for hthgu133pluspm array platform
  6181. \end_layout
  6182. \begin_layout Standard
  6183. In order to enable fRMA normalization for the hthgu133pluspm array platform,
  6184. custom fRMA normalization vectors were trained using the frmaTools package
  6185. \begin_inset CommandInset citation
  6186. LatexCommand cite
  6187. key "McCall2011"
  6188. literal "false"
  6189. \end_inset
  6190. .
  6191. Separate vectors were created for two types of samples: kidney graft biopsy
  6192. samples and blood samples from graft recipients.
  6193. For training, a 341 kidney biopsy samples from 2 data sets and 965 blood
  6194. samples from 5 data sets were used as the reference set.
  6195. Arrays were groups into batches based on unique combinations of sample
  6196. type (blood or biopsy), diagnosis (TX, AR, etc.), data set, and scan date.
  6197. Thus, each batch represents arrays of the same kind that were run together
  6198. on the same day.
  6199. For estimating the probe inverse variance weights, frmaTools requires equal-siz
  6200. ed batches, which means a batch size must be chosen, and then batches smaller
  6201. than that size must be ignored, while batches larger than the chosen size
  6202. must be downsampled.
  6203. This downsampling is performed randomly, so the sampling process is repeated
  6204. 5 times and the resulting normalizations are compared to each other.
  6205. \end_layout
  6206. \begin_layout Standard
  6207. To evaluate the consistency of the generated normalization vectors, the
  6208. 5 fRMA vector sets generated from 5 random batch samplings were each used
  6209. to normalize the same 20 randomly selected samples from each tissue.
  6210. Then the normalized expression values for each probe on each array were
  6211. compared across all normalizations.
  6212. Each fRMA normalization was also compared against the normalized expression
  6213. values obtained by normalizing the same 20 samples with ordinary RMA.
  6214. \end_layout
  6215. \begin_layout Subsection
  6216. Modeling methylation array M-value heteroskedasticy in linear models with
  6217. modified voom implementation
  6218. \end_layout
  6219. \begin_layout Standard
  6220. \begin_inset Flex TODO Note (inline)
  6221. status open
  6222. \begin_layout Plain Layout
  6223. Put code on Github and reference it.
  6224. \end_layout
  6225. \end_inset
  6226. \end_layout
  6227. \begin_layout Standard
  6228. To investigate the whether DNA methylation could be used to distinguish
  6229. between healthy and dysfunctional transplants, a data set of 78 Illumina
  6230. 450k methylation arrays from human kidney graft biopsies was analyzed for
  6231. differential metylation between 4 transplant statuses: healthy transplant
  6232. (TX), transplants undergoing acute rejection (AR), acute dysfunction with
  6233. no rejection (ADNR), and chronic allograpft nephropathy (CAN).
  6234. The data consisted of 33 TX, 9 AR, 8 ADNR, and 28 CAN samples.
  6235. The uneven group sizes are a result of taking the biopsy samples before
  6236. the eventual fate of the transplant was known.
  6237. Each sample was additionally annotated with a donor ID (anonymized), Sex,
  6238. Age, Ethnicity, Creatinine Level, and Diabetes diagnosois (all samples
  6239. in this data set came from patients with either Type 1 or Type 2 diabetes).
  6240. \end_layout
  6241. \begin_layout Standard
  6242. The intensity data were first normalized using subset-quantile within array
  6243. normalization (SWAN)
  6244. \begin_inset CommandInset citation
  6245. LatexCommand cite
  6246. key "Maksimovic2012"
  6247. literal "false"
  6248. \end_inset
  6249. , then converted to intensity ratios (beta values)
  6250. \begin_inset CommandInset citation
  6251. LatexCommand cite
  6252. key "Aryee2014"
  6253. literal "false"
  6254. \end_inset
  6255. .
  6256. Any probes binding to loci that overlapped annotated SNPs were dropped,
  6257. and the annotated sex of each sample was verified against the sex inferred
  6258. from the ratio of median probe intensities for the X and Y chromosomes.
  6259. Then, the ratios were transformed to M-values.
  6260. \end_layout
  6261. \begin_layout Standard
  6262. \begin_inset Float table
  6263. wide false
  6264. sideways false
  6265. status open
  6266. \begin_layout Plain Layout
  6267. \align center
  6268. \begin_inset Tabular
  6269. <lyxtabular version="3" rows="4" columns="6">
  6270. <features tabularvalignment="middle">
  6271. <column alignment="center" valignment="top">
  6272. <column alignment="center" valignment="top">
  6273. <column alignment="center" valignment="top">
  6274. <column alignment="center" valignment="top">
  6275. <column alignment="center" valignment="top">
  6276. <column alignment="center" valignment="top">
  6277. <row>
  6278. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6279. \begin_inset Text
  6280. \begin_layout Plain Layout
  6281. Analysis
  6282. \end_layout
  6283. \end_inset
  6284. </cell>
  6285. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6286. \begin_inset Text
  6287. \begin_layout Plain Layout
  6288. random effect
  6289. \end_layout
  6290. \end_inset
  6291. </cell>
  6292. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6293. \begin_inset Text
  6294. \begin_layout Plain Layout
  6295. eBayes
  6296. \end_layout
  6297. \end_inset
  6298. </cell>
  6299. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6300. \begin_inset Text
  6301. \begin_layout Plain Layout
  6302. SVA
  6303. \end_layout
  6304. \end_inset
  6305. </cell>
  6306. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6307. \begin_inset Text
  6308. \begin_layout Plain Layout
  6309. weights
  6310. \end_layout
  6311. \end_inset
  6312. </cell>
  6313. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  6314. \begin_inset Text
  6315. \begin_layout Plain Layout
  6316. voom
  6317. \end_layout
  6318. \end_inset
  6319. </cell>
  6320. </row>
  6321. <row>
  6322. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6323. \begin_inset Text
  6324. \begin_layout Plain Layout
  6325. A
  6326. \end_layout
  6327. \end_inset
  6328. </cell>
  6329. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6330. \begin_inset Text
  6331. \begin_layout Plain Layout
  6332. Yes
  6333. \end_layout
  6334. \end_inset
  6335. </cell>
  6336. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6337. \begin_inset Text
  6338. \begin_layout Plain Layout
  6339. Yes
  6340. \end_layout
  6341. \end_inset
  6342. </cell>
  6343. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6344. \begin_inset Text
  6345. \begin_layout Plain Layout
  6346. No
  6347. \end_layout
  6348. \end_inset
  6349. </cell>
  6350. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6351. \begin_inset Text
  6352. \begin_layout Plain Layout
  6353. No
  6354. \end_layout
  6355. \end_inset
  6356. </cell>
  6357. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  6358. \begin_inset Text
  6359. \begin_layout Plain Layout
  6360. No
  6361. \end_layout
  6362. \end_inset
  6363. </cell>
  6364. </row>
  6365. <row>
  6366. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6367. \begin_inset Text
  6368. \begin_layout Plain Layout
  6369. B
  6370. \end_layout
  6371. \end_inset
  6372. </cell>
  6373. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6374. \begin_inset Text
  6375. \begin_layout Plain Layout
  6376. Yes
  6377. \end_layout
  6378. \end_inset
  6379. </cell>
  6380. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6381. \begin_inset Text
  6382. \begin_layout Plain Layout
  6383. Yes
  6384. \end_layout
  6385. \end_inset
  6386. </cell>
  6387. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6388. \begin_inset Text
  6389. \begin_layout Plain Layout
  6390. Yes
  6391. \end_layout
  6392. \end_inset
  6393. </cell>
  6394. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6395. \begin_inset Text
  6396. \begin_layout Plain Layout
  6397. Yes
  6398. \end_layout
  6399. \end_inset
  6400. </cell>
  6401. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  6402. \begin_inset Text
  6403. \begin_layout Plain Layout
  6404. No
  6405. \end_layout
  6406. \end_inset
  6407. </cell>
  6408. </row>
  6409. <row>
  6410. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6411. \begin_inset Text
  6412. \begin_layout Plain Layout
  6413. C
  6414. \end_layout
  6415. \end_inset
  6416. </cell>
  6417. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6418. \begin_inset Text
  6419. \begin_layout Plain Layout
  6420. Yes
  6421. \end_layout
  6422. \end_inset
  6423. </cell>
  6424. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6425. \begin_inset Text
  6426. \begin_layout Plain Layout
  6427. Yes
  6428. \end_layout
  6429. \end_inset
  6430. </cell>
  6431. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6432. \begin_inset Text
  6433. \begin_layout Plain Layout
  6434. Yes
  6435. \end_layout
  6436. \end_inset
  6437. </cell>
  6438. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6439. \begin_inset Text
  6440. \begin_layout Plain Layout
  6441. Yes
  6442. \end_layout
  6443. \end_inset
  6444. </cell>
  6445. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  6446. \begin_inset Text
  6447. \begin_layout Plain Layout
  6448. Yes
  6449. \end_layout
  6450. \end_inset
  6451. </cell>
  6452. </row>
  6453. </lyxtabular>
  6454. \end_inset
  6455. \end_layout
  6456. \begin_layout Plain Layout
  6457. \begin_inset Caption Standard
  6458. \begin_layout Plain Layout
  6459. \series bold
  6460. \begin_inset CommandInset label
  6461. LatexCommand label
  6462. name "tab:Summary-of-meth-analysis"
  6463. \end_inset
  6464. Summary of analysis variants for methylation array data.
  6465. \series default
  6466. Each analysis included a different set of steps to adjust or account for
  6467. various systematic features of the data.
  6468. Random effect: The model included a random effect accounting for correlation
  6469. between samples from the same patient
  6470. \begin_inset CommandInset citation
  6471. LatexCommand cite
  6472. key "Smyth2005a"
  6473. literal "false"
  6474. \end_inset
  6475. ; eBayes: Empirical bayes squeezing of per-probe variances toward the mean-varia
  6476. nce trend
  6477. \begin_inset CommandInset citation
  6478. LatexCommand cite
  6479. key "Ritchie2015"
  6480. literal "false"
  6481. \end_inset
  6482. ; SVA: Surrogate variable analysis to account for unobserved confounders
  6483. \begin_inset CommandInset citation
  6484. LatexCommand cite
  6485. key "Leek2007"
  6486. literal "false"
  6487. \end_inset
  6488. ; Weights: Estimate sample weights to account for differences in sample
  6489. quality
  6490. \begin_inset CommandInset citation
  6491. LatexCommand cite
  6492. key "Liu2015,Ritchie2006"
  6493. literal "false"
  6494. \end_inset
  6495. ; voom: Use mean-variance trend to assign individual sample weights
  6496. \begin_inset CommandInset citation
  6497. LatexCommand cite
  6498. key "Law2013"
  6499. literal "false"
  6500. \end_inset
  6501. .
  6502. See the text for a more detailed explanation of each step.
  6503. \end_layout
  6504. \end_inset
  6505. \end_layout
  6506. \end_inset
  6507. \end_layout
  6508. \begin_layout Standard
  6509. From the M-values, a series of parallel analyses was performed, each adding
  6510. additional steps into the model fit to accomodate a feature of the data
  6511. (see Table
  6512. \begin_inset CommandInset ref
  6513. LatexCommand ref
  6514. reference "tab:Summary-of-meth-analysis"
  6515. plural "false"
  6516. caps "false"
  6517. noprefix "false"
  6518. \end_inset
  6519. ).
  6520. For analysis A, a
  6521. \begin_inset Quotes eld
  6522. \end_inset
  6523. basic
  6524. \begin_inset Quotes erd
  6525. \end_inset
  6526. linear modeling analysis was performed, compensating for known confounders
  6527. by including terms for the factor of interest (transplant status) as well
  6528. as the known biological confounders: sex, age, ethnicity, and diabetes.
  6529. Since some samples came from the same patients at different times, the
  6530. intra-patient correlation was modeled as a random effect, estimating a
  6531. shared correlation value across all probes
  6532. \begin_inset CommandInset citation
  6533. LatexCommand cite
  6534. key "Smyth2005a"
  6535. literal "false"
  6536. \end_inset
  6537. .
  6538. Then the linear model was fit, and the variance was modeled using empirical
  6539. Bayes squeezing toward the mean-variance trend
  6540. \begin_inset CommandInset citation
  6541. LatexCommand cite
  6542. key "Ritchie2015"
  6543. literal "false"
  6544. \end_inset
  6545. .
  6546. Finally, t-tests or F-tests were performed as appropriate for each test:
  6547. t-tests for single contrasts, and F-tests for multiple contrasts.
  6548. P-values were corrected for multiple testing using the Benjamini-Hochberg
  6549. procedure for FDR control
  6550. \begin_inset CommandInset citation
  6551. LatexCommand cite
  6552. key "Benjamini1995"
  6553. literal "false"
  6554. \end_inset
  6555. .
  6556. \end_layout
  6557. \begin_layout Standard
  6558. For the analysis B, surrogate variable analysis (SVA) was used to infer
  6559. additional unobserved sources of heterogeneity in the data
  6560. \begin_inset CommandInset citation
  6561. LatexCommand cite
  6562. key "Leek2007"
  6563. literal "false"
  6564. \end_inset
  6565. .
  6566. These surrogate variables were added to the design matrix before fitting
  6567. the linear model.
  6568. In addition, sample quality weights were estimated from the data and used
  6569. during linear modeling to down-weight the contribution of highly variable
  6570. arrays while increasing the weight to arrays with lower variability
  6571. \begin_inset CommandInset citation
  6572. LatexCommand cite
  6573. key "Ritchie2006"
  6574. literal "false"
  6575. \end_inset
  6576. .
  6577. The remainder of the analysis proceeded as in analysis A.
  6578. For analysis C, the voom method was adapted to run on methylation array
  6579. data and used to model and correct for the mean-variance trend using individual
  6580. observation weights
  6581. \begin_inset CommandInset citation
  6582. LatexCommand cite
  6583. key "Law2013"
  6584. literal "false"
  6585. \end_inset
  6586. , which were combined with the sample weights
  6587. \begin_inset CommandInset citation
  6588. LatexCommand cite
  6589. key "Liu2015,Ritchie2006"
  6590. literal "false"
  6591. \end_inset
  6592. .
  6593. Each time weights were used, they were estimated once before estimating
  6594. the random effect correlation value, and then the weights were re-estimated
  6595. taking the random effect into account.
  6596. The remainder of the analysis proceeded as in analysis B.
  6597. \end_layout
  6598. \begin_layout Section
  6599. Results
  6600. \end_layout
  6601. \begin_layout Standard
  6602. \begin_inset Flex TODO Note (inline)
  6603. status open
  6604. \begin_layout Plain Layout
  6605. Improve subsection titles in this section.
  6606. \end_layout
  6607. \end_inset
  6608. \end_layout
  6609. \begin_layout Standard
  6610. \begin_inset Flex TODO Note (inline)
  6611. status open
  6612. \begin_layout Plain Layout
  6613. Reconsider subsection organization?
  6614. \end_layout
  6615. \end_inset
  6616. \end_layout
  6617. \begin_layout Subsection
  6618. Separate normalization with RMA introduces unwanted biases in classification
  6619. \end_layout
  6620. \begin_layout Standard
  6621. \begin_inset Float figure
  6622. wide false
  6623. sideways false
  6624. status open
  6625. \begin_layout Plain Layout
  6626. \align center
  6627. \begin_inset Graphics
  6628. filename graphics/PAM/predplot.pdf
  6629. lyxscale 50
  6630. width 60col%
  6631. groupId colwidth
  6632. \end_inset
  6633. \end_layout
  6634. \begin_layout Plain Layout
  6635. \begin_inset Caption Standard
  6636. \begin_layout Plain Layout
  6637. \begin_inset CommandInset label
  6638. LatexCommand label
  6639. name "fig:Classifier-probabilities-RMA"
  6640. \end_inset
  6641. \series bold
  6642. Classifier probabilities on validation samples when normalized with RMA
  6643. together vs.
  6644. separately.
  6645. \series default
  6646. The PAM classifier algorithm was trained on the training set of arrays to
  6647. distinguish AR from TX and then used to assign class probabilities to the
  6648. validation set.
  6649. The process was performed after normalizing all samples together and after
  6650. normalizing the training and test sets separately, and the class probabilities
  6651. assigned to each sample in the validation set were plotted against each
  6652. other (PP(AR), posterior probability of being AR).
  6653. The color of each point indicates the true classification of that sample.
  6654. \end_layout
  6655. \end_inset
  6656. \end_layout
  6657. \end_inset
  6658. \end_layout
  6659. \begin_layout Standard
  6660. To demonstrate the problem with non-single-channel normalization methods,
  6661. we considered the problem of training a classifier to distinguish TX from
  6662. AR using the samples from the internal set as training data, evaluating
  6663. performance on the external set.
  6664. First, training and evaluation were performed after normalizing all array
  6665. samples together as a single set using RMA, and second, the internal samples
  6666. were normalized separately from the external samples and the training and
  6667. evaluation were repeated.
  6668. For each sample in the validation set, the classifier probabilities from
  6669. both classifiers were plotted against each other (Fig.
  6670. \begin_inset CommandInset ref
  6671. LatexCommand ref
  6672. reference "fig:Classifier-probabilities-RMA"
  6673. plural "false"
  6674. caps "false"
  6675. noprefix "false"
  6676. \end_inset
  6677. ).
  6678. As expected, separate normalization biases the classifier probabilities,
  6679. resulting in several misclassifications.
  6680. In this case, the bias from separate normalization causes the classifier
  6681. to assign a lower probability of AR to every sample.
  6682. \end_layout
  6683. \begin_layout Subsection
  6684. fRMA and SCAN maintain classification performance while eliminating dependence
  6685. on normalization strategy
  6686. \end_layout
  6687. \begin_layout Standard
  6688. \begin_inset Float figure
  6689. wide false
  6690. sideways false
  6691. status open
  6692. \begin_layout Plain Layout
  6693. \align center
  6694. \begin_inset Float figure
  6695. placement tb
  6696. wide false
  6697. sideways false
  6698. status open
  6699. \begin_layout Plain Layout
  6700. \align center
  6701. \begin_inset Graphics
  6702. filename graphics/PAM/ROC-TXvsAR-internal.pdf
  6703. lyxscale 50
  6704. height 40theight%
  6705. groupId roc-pam
  6706. \end_inset
  6707. \end_layout
  6708. \begin_layout Plain Layout
  6709. \begin_inset Caption Standard
  6710. \begin_layout Plain Layout
  6711. \begin_inset CommandInset label
  6712. LatexCommand label
  6713. name "fig:ROC-PAM-int"
  6714. \end_inset
  6715. ROC curves for PAM on internal validation data
  6716. \end_layout
  6717. \end_inset
  6718. \end_layout
  6719. \end_inset
  6720. \end_layout
  6721. \begin_layout Plain Layout
  6722. \align center
  6723. \begin_inset Float figure
  6724. placement tb
  6725. wide false
  6726. sideways false
  6727. status open
  6728. \begin_layout Plain Layout
  6729. \align center
  6730. \begin_inset Graphics
  6731. filename graphics/PAM/ROC-TXvsAR-external.pdf
  6732. lyxscale 50
  6733. height 40theight%
  6734. groupId roc-pam
  6735. \end_inset
  6736. \end_layout
  6737. \begin_layout Plain Layout
  6738. \begin_inset Caption Standard
  6739. \begin_layout Plain Layout
  6740. \begin_inset CommandInset label
  6741. LatexCommand label
  6742. name "fig:ROC-PAM-ext"
  6743. \end_inset
  6744. ROC curves for PAM on external validation data
  6745. \end_layout
  6746. \end_inset
  6747. \end_layout
  6748. \end_inset
  6749. \end_layout
  6750. \begin_layout Plain Layout
  6751. \begin_inset Caption Standard
  6752. \begin_layout Plain Layout
  6753. \series bold
  6754. \begin_inset CommandInset label
  6755. LatexCommand label
  6756. name "fig:ROC-PAM-main"
  6757. \end_inset
  6758. ROC curves for PAM using different normalization strategies.
  6759. \series default
  6760. ROC curves were generated for PAM classification of AR vs TX after 6 different
  6761. normalization strategies applied to the same data sets.
  6762. Only fRMA and SCAN are single-channel normalizations.
  6763. The other normalizations are for comparison.
  6764. \end_layout
  6765. \end_inset
  6766. \end_layout
  6767. \end_inset
  6768. \end_layout
  6769. \begin_layout Standard
  6770. \begin_inset Float table
  6771. wide false
  6772. sideways false
  6773. status open
  6774. \begin_layout Plain Layout
  6775. \align center
  6776. \begin_inset Tabular
  6777. <lyxtabular version="3" rows="7" columns="4">
  6778. <features tabularvalignment="middle">
  6779. <column alignment="center" valignment="top">
  6780. <column alignment="center" valignment="top">
  6781. <column alignment="center" valignment="top">
  6782. <column alignment="center" valignment="top">
  6783. <row>
  6784. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6785. \begin_inset Text
  6786. \begin_layout Plain Layout
  6787. \family roman
  6788. \series medium
  6789. \shape up
  6790. \size normal
  6791. \emph off
  6792. \bar no
  6793. \strikeout off
  6794. \xout off
  6795. \uuline off
  6796. \uwave off
  6797. \noun off
  6798. \color none
  6799. Normalization
  6800. \end_layout
  6801. \end_inset
  6802. </cell>
  6803. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6804. \begin_inset Text
  6805. \begin_layout Plain Layout
  6806. Single-channel?
  6807. \end_layout
  6808. \end_inset
  6809. </cell>
  6810. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  6811. \begin_inset Text
  6812. \begin_layout Plain Layout
  6813. \family roman
  6814. \series medium
  6815. \shape up
  6816. \size normal
  6817. \emph off
  6818. \bar no
  6819. \strikeout off
  6820. \xout off
  6821. \uuline off
  6822. \uwave off
  6823. \noun off
  6824. \color none
  6825. Internal Val.
  6826. AUC
  6827. \end_layout
  6828. \end_inset
  6829. </cell>
  6830. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  6831. \begin_inset Text
  6832. \begin_layout Plain Layout
  6833. External Val.
  6834. AUC
  6835. \end_layout
  6836. \end_inset
  6837. </cell>
  6838. </row>
  6839. <row>
  6840. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6841. \begin_inset Text
  6842. \begin_layout Plain Layout
  6843. \family roman
  6844. \series medium
  6845. \shape up
  6846. \size normal
  6847. \emph off
  6848. \bar no
  6849. \strikeout off
  6850. \xout off
  6851. \uuline off
  6852. \uwave off
  6853. \noun off
  6854. \color none
  6855. RMA
  6856. \end_layout
  6857. \end_inset
  6858. </cell>
  6859. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6860. \begin_inset Text
  6861. \begin_layout Plain Layout
  6862. No
  6863. \end_layout
  6864. \end_inset
  6865. </cell>
  6866. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6867. \begin_inset Text
  6868. \begin_layout Plain Layout
  6869. \family roman
  6870. \series medium
  6871. \shape up
  6872. \size normal
  6873. \emph off
  6874. \bar no
  6875. \strikeout off
  6876. \xout off
  6877. \uuline off
  6878. \uwave off
  6879. \noun off
  6880. \color none
  6881. 0.852
  6882. \end_layout
  6883. \end_inset
  6884. </cell>
  6885. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  6886. \begin_inset Text
  6887. \begin_layout Plain Layout
  6888. \family roman
  6889. \series medium
  6890. \shape up
  6891. \size normal
  6892. \emph off
  6893. \bar no
  6894. \strikeout off
  6895. \xout off
  6896. \uuline off
  6897. \uwave off
  6898. \noun off
  6899. \color none
  6900. 0.713
  6901. \end_layout
  6902. \end_inset
  6903. </cell>
  6904. </row>
  6905. <row>
  6906. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6907. \begin_inset Text
  6908. \begin_layout Plain Layout
  6909. \family roman
  6910. \series medium
  6911. \shape up
  6912. \size normal
  6913. \emph off
  6914. \bar no
  6915. \strikeout off
  6916. \xout off
  6917. \uuline off
  6918. \uwave off
  6919. \noun off
  6920. \color none
  6921. dChip
  6922. \end_layout
  6923. \end_inset
  6924. </cell>
  6925. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6926. \begin_inset Text
  6927. \begin_layout Plain Layout
  6928. No
  6929. \end_layout
  6930. \end_inset
  6931. </cell>
  6932. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6933. \begin_inset Text
  6934. \begin_layout Plain Layout
  6935. \family roman
  6936. \series medium
  6937. \shape up
  6938. \size normal
  6939. \emph off
  6940. \bar no
  6941. \strikeout off
  6942. \xout off
  6943. \uuline off
  6944. \uwave off
  6945. \noun off
  6946. \color none
  6947. 0.891
  6948. \end_layout
  6949. \end_inset
  6950. </cell>
  6951. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  6952. \begin_inset Text
  6953. \begin_layout Plain Layout
  6954. \family roman
  6955. \series medium
  6956. \shape up
  6957. \size normal
  6958. \emph off
  6959. \bar no
  6960. \strikeout off
  6961. \xout off
  6962. \uuline off
  6963. \uwave off
  6964. \noun off
  6965. \color none
  6966. 0.657
  6967. \end_layout
  6968. \end_inset
  6969. </cell>
  6970. </row>
  6971. <row>
  6972. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6973. \begin_inset Text
  6974. \begin_layout Plain Layout
  6975. \family roman
  6976. \series medium
  6977. \shape up
  6978. \size normal
  6979. \emph off
  6980. \bar no
  6981. \strikeout off
  6982. \xout off
  6983. \uuline off
  6984. \uwave off
  6985. \noun off
  6986. \color none
  6987. RMA + GRSN
  6988. \end_layout
  6989. \end_inset
  6990. </cell>
  6991. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6992. \begin_inset Text
  6993. \begin_layout Plain Layout
  6994. No
  6995. \end_layout
  6996. \end_inset
  6997. </cell>
  6998. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  6999. \begin_inset Text
  7000. \begin_layout Plain Layout
  7001. \family roman
  7002. \series medium
  7003. \shape up
  7004. \size normal
  7005. \emph off
  7006. \bar no
  7007. \strikeout off
  7008. \xout off
  7009. \uuline off
  7010. \uwave off
  7011. \noun off
  7012. \color none
  7013. 0.816
  7014. \end_layout
  7015. \end_inset
  7016. </cell>
  7017. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  7018. \begin_inset Text
  7019. \begin_layout Plain Layout
  7020. \family roman
  7021. \series medium
  7022. \shape up
  7023. \size normal
  7024. \emph off
  7025. \bar no
  7026. \strikeout off
  7027. \xout off
  7028. \uuline off
  7029. \uwave off
  7030. \noun off
  7031. \color none
  7032. 0.750
  7033. \end_layout
  7034. \end_inset
  7035. </cell>
  7036. </row>
  7037. <row>
  7038. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  7039. \begin_inset Text
  7040. \begin_layout Plain Layout
  7041. \family roman
  7042. \series medium
  7043. \shape up
  7044. \size normal
  7045. \emph off
  7046. \bar no
  7047. \strikeout off
  7048. \xout off
  7049. \uuline off
  7050. \uwave off
  7051. \noun off
  7052. \color none
  7053. dChip + GRSN
  7054. \end_layout
  7055. \end_inset
  7056. </cell>
  7057. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  7058. \begin_inset Text
  7059. \begin_layout Plain Layout
  7060. No
  7061. \end_layout
  7062. \end_inset
  7063. </cell>
  7064. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  7065. \begin_inset Text
  7066. \begin_layout Plain Layout
  7067. \family roman
  7068. \series medium
  7069. \shape up
  7070. \size normal
  7071. \emph off
  7072. \bar no
  7073. \strikeout off
  7074. \xout off
  7075. \uuline off
  7076. \uwave off
  7077. \noun off
  7078. \color none
  7079. 0.875
  7080. \end_layout
  7081. \end_inset
  7082. </cell>
  7083. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  7084. \begin_inset Text
  7085. \begin_layout Plain Layout
  7086. \family roman
  7087. \series medium
  7088. \shape up
  7089. \size normal
  7090. \emph off
  7091. \bar no
  7092. \strikeout off
  7093. \xout off
  7094. \uuline off
  7095. \uwave off
  7096. \noun off
  7097. \color none
  7098. 0.642
  7099. \end_layout
  7100. \end_inset
  7101. </cell>
  7102. </row>
  7103. <row>
  7104. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  7105. \begin_inset Text
  7106. \begin_layout Plain Layout
  7107. \family roman
  7108. \series medium
  7109. \shape up
  7110. \size normal
  7111. \emph off
  7112. \bar no
  7113. \strikeout off
  7114. \xout off
  7115. \uuline off
  7116. \uwave off
  7117. \noun off
  7118. \color none
  7119. fRMA
  7120. \end_layout
  7121. \end_inset
  7122. </cell>
  7123. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  7124. \begin_inset Text
  7125. \begin_layout Plain Layout
  7126. Yes
  7127. \end_layout
  7128. \end_inset
  7129. </cell>
  7130. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  7131. \begin_inset Text
  7132. \begin_layout Plain Layout
  7133. \family roman
  7134. \series medium
  7135. \shape up
  7136. \size normal
  7137. \emph off
  7138. \bar no
  7139. \strikeout off
  7140. \xout off
  7141. \uuline off
  7142. \uwave off
  7143. \noun off
  7144. \color none
  7145. 0.863
  7146. \end_layout
  7147. \end_inset
  7148. </cell>
  7149. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  7150. \begin_inset Text
  7151. \begin_layout Plain Layout
  7152. \family roman
  7153. \series medium
  7154. \shape up
  7155. \size normal
  7156. \emph off
  7157. \bar no
  7158. \strikeout off
  7159. \xout off
  7160. \uuline off
  7161. \uwave off
  7162. \noun off
  7163. \color none
  7164. 0.718
  7165. \end_layout
  7166. \end_inset
  7167. </cell>
  7168. </row>
  7169. <row>
  7170. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  7171. \begin_inset Text
  7172. \begin_layout Plain Layout
  7173. \family roman
  7174. \series medium
  7175. \shape up
  7176. \size normal
  7177. \emph off
  7178. \bar no
  7179. \strikeout off
  7180. \xout off
  7181. \uuline off
  7182. \uwave off
  7183. \noun off
  7184. \color none
  7185. SCAN
  7186. \end_layout
  7187. \end_inset
  7188. </cell>
  7189. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  7190. \begin_inset Text
  7191. \begin_layout Plain Layout
  7192. Yes
  7193. \end_layout
  7194. \end_inset
  7195. </cell>
  7196. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  7197. \begin_inset Text
  7198. \begin_layout Plain Layout
  7199. \family roman
  7200. \series medium
  7201. \shape up
  7202. \size normal
  7203. \emph off
  7204. \bar no
  7205. \strikeout off
  7206. \xout off
  7207. \uuline off
  7208. \uwave off
  7209. \noun off
  7210. \color none
  7211. 0.853
  7212. \end_layout
  7213. \end_inset
  7214. </cell>
  7215. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  7216. \begin_inset Text
  7217. \begin_layout Plain Layout
  7218. \family roman
  7219. \series medium
  7220. \shape up
  7221. \size normal
  7222. \emph off
  7223. \bar no
  7224. \strikeout off
  7225. \xout off
  7226. \uuline off
  7227. \uwave off
  7228. \noun off
  7229. \color none
  7230. 0.689
  7231. \end_layout
  7232. \end_inset
  7233. </cell>
  7234. </row>
  7235. </lyxtabular>
  7236. \end_inset
  7237. \end_layout
  7238. \begin_layout Plain Layout
  7239. \begin_inset Caption Standard
  7240. \begin_layout Plain Layout
  7241. \begin_inset CommandInset label
  7242. LatexCommand label
  7243. name "tab:AUC-PAM"
  7244. \end_inset
  7245. \series bold
  7246. ROC curve AUC values for internal and external validation with 6 different
  7247. normalization strategies.
  7248. \series default
  7249. These AUC values correspond to the ROC curves in Figure
  7250. \begin_inset CommandInset ref
  7251. LatexCommand ref
  7252. reference "fig:ROC-PAM-main"
  7253. plural "false"
  7254. caps "false"
  7255. noprefix "false"
  7256. \end_inset
  7257. .
  7258. \end_layout
  7259. \end_inset
  7260. \end_layout
  7261. \end_inset
  7262. \end_layout
  7263. \begin_layout Standard
  7264. For internal validation, the 6 methods' AUC values ranged from 0.816 to 0.891,
  7265. as shown in Table
  7266. \begin_inset CommandInset ref
  7267. LatexCommand ref
  7268. reference "tab:AUC-PAM"
  7269. plural "false"
  7270. caps "false"
  7271. noprefix "false"
  7272. \end_inset
  7273. .
  7274. Among the non-single-channel normalizations, dChip outperformed RMA, while
  7275. GRSN reduced the AUC values for both dChip and RMA.
  7276. Both single-channel methods, fRMA and SCAN, slightly outperformed RMA,
  7277. with fRMA ahead of SCAN.
  7278. However, the difference between RMA and fRMA is still quite small.
  7279. Figure
  7280. \begin_inset CommandInset ref
  7281. LatexCommand ref
  7282. reference "fig:ROC-PAM-int"
  7283. plural "false"
  7284. caps "false"
  7285. noprefix "false"
  7286. \end_inset
  7287. shows that the ROC curves for RMA, dChip, and fRMA look very similar and
  7288. relatively smooth, while both GRSN curves and the curve for SCAN have a
  7289. more jagged appearance.
  7290. \end_layout
  7291. \begin_layout Standard
  7292. For external validation, as expected, all the AUC values are lower than
  7293. the internal validations, ranging from 0.642 to 0.750 (Table
  7294. \begin_inset CommandInset ref
  7295. LatexCommand ref
  7296. reference "tab:AUC-PAM"
  7297. plural "false"
  7298. caps "false"
  7299. noprefix "false"
  7300. \end_inset
  7301. ).
  7302. With or without GRSN, RMA shows its dominance over dChip in this more challengi
  7303. ng test.
  7304. Unlike in the internal validation, GRSN actually improves the classifier
  7305. performance for RMA, although it does not for dChip.
  7306. Once again, both single-channel methods perform about on par with RMA,
  7307. with fRMA performing slightly better and SCAN performing a bit worse.
  7308. Figure
  7309. \begin_inset CommandInset ref
  7310. LatexCommand ref
  7311. reference "fig:ROC-PAM-ext"
  7312. plural "false"
  7313. caps "false"
  7314. noprefix "false"
  7315. \end_inset
  7316. shows the ROC curves for the external validation test.
  7317. As expected, none of them are as clean-looking as the internal validation
  7318. ROC curves.
  7319. The curves for RMA, RMA+GRSN, and fRMA all look similar, while the other
  7320. curves look more divergent.
  7321. \end_layout
  7322. \begin_layout Subsection
  7323. fRMA with custom-generated vectors enables single-channel normalization
  7324. on hthgu133pluspm platform
  7325. \end_layout
  7326. \begin_layout Standard
  7327. \begin_inset Float figure
  7328. wide false
  7329. sideways false
  7330. status open
  7331. \begin_layout Plain Layout
  7332. \align center
  7333. \begin_inset Float figure
  7334. placement tb
  7335. wide false
  7336. sideways false
  7337. status collapsed
  7338. \begin_layout Plain Layout
  7339. \align center
  7340. \begin_inset Graphics
  7341. filename graphics/frma-pax-bx/batchsize_batches.pdf
  7342. lyxscale 50
  7343. height 35theight%
  7344. groupId frmatools-subfig
  7345. \end_inset
  7346. \end_layout
  7347. \begin_layout Plain Layout
  7348. \begin_inset Caption Standard
  7349. \begin_layout Plain Layout
  7350. \begin_inset CommandInset label
  7351. LatexCommand label
  7352. name "fig:batch-size-batches"
  7353. \end_inset
  7354. \series bold
  7355. Number of batches usable in fRMA probe weight learning as a function of
  7356. batch size.
  7357. \end_layout
  7358. \end_inset
  7359. \end_layout
  7360. \end_inset
  7361. \end_layout
  7362. \begin_layout Plain Layout
  7363. \align center
  7364. \begin_inset Float figure
  7365. placement tb
  7366. wide false
  7367. sideways false
  7368. status collapsed
  7369. \begin_layout Plain Layout
  7370. \align center
  7371. \begin_inset Graphics
  7372. filename graphics/frma-pax-bx/batchsize_samples.pdf
  7373. lyxscale 50
  7374. height 35theight%
  7375. groupId frmatools-subfig
  7376. \end_inset
  7377. \end_layout
  7378. \begin_layout Plain Layout
  7379. \begin_inset Caption Standard
  7380. \begin_layout Plain Layout
  7381. \begin_inset CommandInset label
  7382. LatexCommand label
  7383. name "fig:batch-size-samples"
  7384. \end_inset
  7385. \series bold
  7386. Number of samples usable in fRMA probe weight learning as a function of
  7387. batch size.
  7388. \end_layout
  7389. \end_inset
  7390. \end_layout
  7391. \end_inset
  7392. \end_layout
  7393. \begin_layout Plain Layout
  7394. \begin_inset Caption Standard
  7395. \begin_layout Plain Layout
  7396. \series bold
  7397. \begin_inset CommandInset label
  7398. LatexCommand label
  7399. name "fig:frmatools-batch-size"
  7400. \end_inset
  7401. Effect of batch size selection on number of batches and number of samples
  7402. included in fRMA probe weight learning.
  7403. \series default
  7404. For batch sizes ranging from 3 to 15, the number of batches (a) and samples
  7405. (b) included in probe weight training were plotted for biopsy (BX) and
  7406. blood (PAX) samples.
  7407. The selected batch size, 5, is marked with a dotted vertical line.
  7408. \end_layout
  7409. \end_inset
  7410. \end_layout
  7411. \end_inset
  7412. \end_layout
  7413. \begin_layout Standard
  7414. In order to enable use of fRMA to normalize hthgu133pluspm, a custom set
  7415. of fRMA vectors was created.
  7416. First, an appropriate batch size was chosen by looking at the number of
  7417. batches and number of samples included as a function of batch size (Figure
  7418. \begin_inset CommandInset ref
  7419. LatexCommand ref
  7420. reference "fig:frmatools-batch-size"
  7421. plural "false"
  7422. caps "false"
  7423. noprefix "false"
  7424. \end_inset
  7425. ).
  7426. For a given batch size, all batches with fewer samples that the chosen
  7427. size must be ignored during training, while larger batches must be randomly
  7428. downsampled to the chosen size.
  7429. Hence, the number of samples included for a given batch size equals the
  7430. batch size times the number of batches with at least that many samples.
  7431. From Figure
  7432. \begin_inset CommandInset ref
  7433. LatexCommand ref
  7434. reference "fig:batch-size-samples"
  7435. plural "false"
  7436. caps "false"
  7437. noprefix "false"
  7438. \end_inset
  7439. , it is apparent that that a batch size of 8 maximizes the number of samples
  7440. included in training.
  7441. Increasing the batch size beyond this causes too many smaller batches to
  7442. be excluded, reducing the total number of samples for both tissue types.
  7443. However, a batch size of 8 is not necessarily optimal.
  7444. The article introducing frmaTools concluded that it was highly advantageous
  7445. to use a smaller batch size in order to include more batches, even at the
  7446. expense of including fewer total samples in training
  7447. \begin_inset CommandInset citation
  7448. LatexCommand cite
  7449. key "McCall2011"
  7450. literal "false"
  7451. \end_inset
  7452. .
  7453. To strike an appropriate balance between more batches and more samples,
  7454. a batch size of 5 was chosen.
  7455. For both blood and biopsy samples, this increased the number of batches
  7456. included by 10, with only a modest reduction in the number of samples compared
  7457. to a batch size of 8.
  7458. With a batch size of 5, 26 batches of biopsy samples and 46 batches of
  7459. blood samples were available.
  7460. \end_layout
  7461. \begin_layout Standard
  7462. \begin_inset Float figure
  7463. wide false
  7464. sideways false
  7465. status collapsed
  7466. \begin_layout Plain Layout
  7467. \begin_inset Float figure
  7468. wide false
  7469. sideways false
  7470. status open
  7471. \begin_layout Plain Layout
  7472. \align center
  7473. \begin_inset Graphics
  7474. filename graphics/frma-pax-bx/M-BX-violin.pdf
  7475. lyxscale 40
  7476. width 45col%
  7477. groupId m-violin
  7478. \end_inset
  7479. \end_layout
  7480. \begin_layout Plain Layout
  7481. \begin_inset Caption Standard
  7482. \begin_layout Plain Layout
  7483. \begin_inset CommandInset label
  7484. LatexCommand label
  7485. name "fig:m-bx-violin"
  7486. \end_inset
  7487. \series bold
  7488. Violin plot of inter-normalization log ratios for biopsy samples.
  7489. \end_layout
  7490. \end_inset
  7491. \end_layout
  7492. \end_inset
  7493. \begin_inset space \hfill{}
  7494. \end_inset
  7495. \begin_inset Float figure
  7496. wide false
  7497. sideways false
  7498. status collapsed
  7499. \begin_layout Plain Layout
  7500. \align center
  7501. \begin_inset Graphics
  7502. filename graphics/frma-pax-bx/M-PAX-violin.pdf
  7503. lyxscale 40
  7504. width 45col%
  7505. groupId m-violin
  7506. \end_inset
  7507. \end_layout
  7508. \begin_layout Plain Layout
  7509. \begin_inset Caption Standard
  7510. \begin_layout Plain Layout
  7511. \begin_inset CommandInset label
  7512. LatexCommand label
  7513. name "fig:m-pax-violin"
  7514. \end_inset
  7515. \series bold
  7516. Violin plot of inter-normalization log ratios for blood samples.
  7517. \end_layout
  7518. \end_inset
  7519. \end_layout
  7520. \end_inset
  7521. \end_layout
  7522. \begin_layout Plain Layout
  7523. \begin_inset Caption Standard
  7524. \begin_layout Plain Layout
  7525. \begin_inset CommandInset label
  7526. LatexCommand label
  7527. name "fig:frma-violin"
  7528. \end_inset
  7529. \series bold
  7530. Violin plot of log ratios between normalizations for 20 biopsy samples.
  7531. \series default
  7532. Each of 20 randomly selected samples was normalized with RMA and with 5
  7533. different sets of fRMA vectors.
  7534. The distribution of log ratios between normalized expression values, aggregated
  7535. across all 20 arrays, was plotted for each pair of normalizations.
  7536. \end_layout
  7537. \end_inset
  7538. \end_layout
  7539. \end_inset
  7540. \end_layout
  7541. \begin_layout Standard
  7542. Since fRMA training requires equal-size batches, larger batches are downsampled
  7543. randomly.
  7544. This introduces a nondeterministic step in the generation of normalization
  7545. vectors.
  7546. To show that this randomness does not substantially change the outcome,
  7547. the random downsampling and subsequent vector learning was repeated 5 times,
  7548. with a different random seed each time.
  7549. 20 samples were selected at random as a test set and normalized with each
  7550. of the 5 sets of fRMA normalization vectors as well as ordinary RMA, and
  7551. the normalized expression values were compared across normalizations.
  7552. Figure
  7553. \begin_inset CommandInset ref
  7554. LatexCommand ref
  7555. reference "fig:m-bx-violin"
  7556. plural "false"
  7557. caps "false"
  7558. noprefix "false"
  7559. \end_inset
  7560. shows a summary of these comparisons for biopsy samples.
  7561. Comparing RMA to each of the 5 fRMA normalizations, the distribution of
  7562. log ratios is somewhat wide, indicating that the normalizations disagree
  7563. on the expression values of a fair number of probe sets.
  7564. In contrast, comparisons of fRMA against fRMA, the vast mojority of probe
  7565. sets have very small log ratios, indicating a very high agreement between
  7566. the normalized values generated by the two normalizations.
  7567. This shows that the fRMA normalization's behavior is not very sensitive
  7568. to the random downsampling of larger batches during training.
  7569. \end_layout
  7570. \begin_layout Standard
  7571. \begin_inset Float figure
  7572. wide false
  7573. sideways false
  7574. status open
  7575. \begin_layout Plain Layout
  7576. \align center
  7577. \begin_inset Float figure
  7578. wide false
  7579. sideways false
  7580. status collapsed
  7581. \begin_layout Plain Layout
  7582. \align center
  7583. \begin_inset Graphics
  7584. filename graphics/frma-pax-bx/MA-BX-RMA.fRMA-RASTER.png
  7585. lyxscale 10
  7586. width 45col%
  7587. groupId ma-frma
  7588. \end_inset
  7589. \end_layout
  7590. \begin_layout Plain Layout
  7591. \begin_inset Caption Standard
  7592. \begin_layout Plain Layout
  7593. \begin_inset CommandInset label
  7594. LatexCommand label
  7595. name "fig:ma-bx-rma-frma"
  7596. \end_inset
  7597. RMA vs.
  7598. fRMA for biopsy samples.
  7599. \end_layout
  7600. \end_inset
  7601. \end_layout
  7602. \end_inset
  7603. \begin_inset space \hfill{}
  7604. \end_inset
  7605. \begin_inset Float figure
  7606. wide false
  7607. sideways false
  7608. status collapsed
  7609. \begin_layout Plain Layout
  7610. \align center
  7611. \begin_inset Graphics
  7612. filename graphics/frma-pax-bx/MA-BX-fRMA.fRMA-RASTER.png
  7613. lyxscale 10
  7614. width 45col%
  7615. groupId ma-frma
  7616. \end_inset
  7617. \end_layout
  7618. \begin_layout Plain Layout
  7619. \begin_inset Caption Standard
  7620. \begin_layout Plain Layout
  7621. \begin_inset CommandInset label
  7622. LatexCommand label
  7623. name "fig:ma-bx-frma-frma"
  7624. \end_inset
  7625. fRMA vs fRMA for biopsy samples.
  7626. \end_layout
  7627. \end_inset
  7628. \end_layout
  7629. \end_inset
  7630. \end_layout
  7631. \begin_layout Plain Layout
  7632. \align center
  7633. \begin_inset Float figure
  7634. wide false
  7635. sideways false
  7636. status collapsed
  7637. \begin_layout Plain Layout
  7638. \align center
  7639. \begin_inset Graphics
  7640. filename graphics/frma-pax-bx/MA-PAX-RMA.fRMA-RASTER.png
  7641. lyxscale 10
  7642. width 45col%
  7643. groupId ma-frma
  7644. \end_inset
  7645. \end_layout
  7646. \begin_layout Plain Layout
  7647. \begin_inset Caption Standard
  7648. \begin_layout Plain Layout
  7649. \begin_inset CommandInset label
  7650. LatexCommand label
  7651. name "fig:MA-PAX-rma-frma"
  7652. \end_inset
  7653. RMA vs.
  7654. fRMA for blood samples.
  7655. \end_layout
  7656. \end_inset
  7657. \end_layout
  7658. \end_inset
  7659. \begin_inset space \hfill{}
  7660. \end_inset
  7661. \begin_inset Float figure
  7662. wide false
  7663. sideways false
  7664. status collapsed
  7665. \begin_layout Plain Layout
  7666. \align center
  7667. \begin_inset Graphics
  7668. filename graphics/frma-pax-bx/MA-PAX-fRMA.fRMA-RASTER.png
  7669. lyxscale 10
  7670. width 45col%
  7671. groupId ma-frma
  7672. \end_inset
  7673. \end_layout
  7674. \begin_layout Plain Layout
  7675. \begin_inset Caption Standard
  7676. \begin_layout Plain Layout
  7677. \begin_inset CommandInset label
  7678. LatexCommand label
  7679. name "fig:MA-PAX-frma-frma"
  7680. \end_inset
  7681. fRMA vs fRMA for blood samples.
  7682. \end_layout
  7683. \end_inset
  7684. \end_layout
  7685. \end_inset
  7686. \end_layout
  7687. \begin_layout Plain Layout
  7688. \begin_inset Caption Standard
  7689. \begin_layout Plain Layout
  7690. \series bold
  7691. \begin_inset CommandInset label
  7692. LatexCommand label
  7693. name "fig:Representative-MA-plots"
  7694. \end_inset
  7695. Representative MA plots comparing RMA and custom fRMA normalizations.
  7696. \series default
  7697. For each plot, 20 samples were normalized using 2 different normalizations,
  7698. and then averages (A) and log ratios (M) were plotted between the two different
  7699. normalizations for every probe.
  7700. For the
  7701. \begin_inset Quotes eld
  7702. \end_inset
  7703. fRMA vs fRMA
  7704. \begin_inset Quotes erd
  7705. \end_inset
  7706. plots (b & d), two different fRMA normalizations using vectors from two
  7707. independent batch samplings were compared.
  7708. Density of points is represented by blue shading, and individual outlier
  7709. points are plotted.
  7710. \end_layout
  7711. \end_inset
  7712. \end_layout
  7713. \end_inset
  7714. \end_layout
  7715. \begin_layout Standard
  7716. Figure
  7717. \begin_inset CommandInset ref
  7718. LatexCommand ref
  7719. reference "fig:ma-bx-rma-frma"
  7720. plural "false"
  7721. caps "false"
  7722. noprefix "false"
  7723. \end_inset
  7724. shows an MA plot of the RMA-normalized values against the fRMA-normalized
  7725. values for the same probe sets and arrays, corresponding to the first row
  7726. of Figure
  7727. \begin_inset CommandInset ref
  7728. LatexCommand ref
  7729. reference "fig:m-bx-violin"
  7730. plural "false"
  7731. caps "false"
  7732. noprefix "false"
  7733. \end_inset
  7734. .
  7735. This MA plot shows that not only is there a wide distribution of M-values,
  7736. but the trend of M-values is dependent on the average normalized intensity.
  7737. This is expected, since the overall trend represents the differences in
  7738. the quantile normalization step.
  7739. When running RMA, only the quantiles for these specific 20 arrays are used,
  7740. while for fRMA the quantile distribution is taking from all arrays used
  7741. in training.
  7742. Figure
  7743. \begin_inset CommandInset ref
  7744. LatexCommand ref
  7745. reference "fig:ma-bx-frma-frma"
  7746. plural "false"
  7747. caps "false"
  7748. noprefix "false"
  7749. \end_inset
  7750. shows a similar MA plot comparing 2 different fRMA normalizations, correspondin
  7751. g to the 6th row of Figure
  7752. \begin_inset CommandInset ref
  7753. LatexCommand ref
  7754. reference "fig:m-bx-violin"
  7755. plural "false"
  7756. caps "false"
  7757. noprefix "false"
  7758. \end_inset
  7759. .
  7760. The MA plot is very tightly centered around zero with no visible trend.
  7761. Figures
  7762. \begin_inset CommandInset ref
  7763. LatexCommand ref
  7764. reference "fig:m-pax-violin"
  7765. plural "false"
  7766. caps "false"
  7767. noprefix "false"
  7768. \end_inset
  7769. ,
  7770. \begin_inset CommandInset ref
  7771. LatexCommand ref
  7772. reference "fig:MA-PAX-rma-frma"
  7773. plural "false"
  7774. caps "false"
  7775. noprefix "false"
  7776. \end_inset
  7777. , and
  7778. \begin_inset CommandInset ref
  7779. LatexCommand ref
  7780. reference "fig:ma-bx-frma-frma"
  7781. plural "false"
  7782. caps "false"
  7783. noprefix "false"
  7784. \end_inset
  7785. show exactly the same information for the blood samples, once again comparing
  7786. the normalized expression values between normalizations for all probe sets
  7787. across 20 randomly selected test arrays.
  7788. Once again, there is a wider distribution of log ratios between RMA-normalized
  7789. values and fRMA-normalized, and a much tighter distribution when comparing
  7790. different fRMA normalizations to each other, indicating that the fRMA training
  7791. process is robust to random batch downsampling for the blood samples as
  7792. well.
  7793. \end_layout
  7794. \begin_layout Subsection
  7795. SVA, voom, and array weights improve model fit for methylation array data
  7796. \end_layout
  7797. \begin_layout Standard
  7798. \begin_inset ERT
  7799. status open
  7800. \begin_layout Plain Layout
  7801. \backslash
  7802. afterpage{
  7803. \end_layout
  7804. \begin_layout Plain Layout
  7805. \backslash
  7806. begin{landscape}
  7807. \end_layout
  7808. \end_inset
  7809. \end_layout
  7810. \begin_layout Standard
  7811. \begin_inset Float figure
  7812. wide false
  7813. sideways false
  7814. status open
  7815. \begin_layout Plain Layout
  7816. \begin_inset Flex TODO Note (inline)
  7817. status open
  7818. \begin_layout Plain Layout
  7819. Fix axis labels:
  7820. \begin_inset Quotes eld
  7821. \end_inset
  7822. log2 M-value
  7823. \begin_inset Quotes erd
  7824. \end_inset
  7825. is redundant because M-values are already log scale
  7826. \end_layout
  7827. \end_inset
  7828. \end_layout
  7829. \begin_layout Plain Layout
  7830. \begin_inset Float figure
  7831. wide false
  7832. sideways false
  7833. status collapsed
  7834. \begin_layout Plain Layout
  7835. \align center
  7836. \begin_inset Graphics
  7837. filename graphics/methylvoom/unadj.dupcor/meanvar-trends-PAGE1-CROP-RASTER.png
  7838. lyxscale 15
  7839. width 30col%
  7840. groupId voomaw-subfig
  7841. \end_inset
  7842. \end_layout
  7843. \begin_layout Plain Layout
  7844. \begin_inset Caption Standard
  7845. \begin_layout Plain Layout
  7846. \begin_inset CommandInset label
  7847. LatexCommand label
  7848. name "fig:meanvar-basic"
  7849. \end_inset
  7850. Mean-variance trend for analysis A.
  7851. \end_layout
  7852. \end_inset
  7853. \end_layout
  7854. \end_inset
  7855. \begin_inset space \hfill{}
  7856. \end_inset
  7857. \begin_inset Float figure
  7858. wide false
  7859. sideways false
  7860. status collapsed
  7861. \begin_layout Plain Layout
  7862. \align center
  7863. \begin_inset Graphics
  7864. filename graphics/methylvoom/unadj.dupcor.sva.aw/meanvar-trends-PAGE1-CROP-RASTER.png
  7865. lyxscale 15
  7866. width 30col%
  7867. groupId voomaw-subfig
  7868. \end_inset
  7869. \end_layout
  7870. \begin_layout Plain Layout
  7871. \begin_inset Caption Standard
  7872. \begin_layout Plain Layout
  7873. \begin_inset CommandInset label
  7874. LatexCommand label
  7875. name "fig:meanvar-sva-aw"
  7876. \end_inset
  7877. Mean-variance trend for analysis B.
  7878. \end_layout
  7879. \end_inset
  7880. \end_layout
  7881. \end_inset
  7882. \begin_inset space \hfill{}
  7883. \end_inset
  7884. \begin_inset Float figure
  7885. wide false
  7886. sideways false
  7887. status collapsed
  7888. \begin_layout Plain Layout
  7889. \align center
  7890. \begin_inset Graphics
  7891. filename graphics/methylvoom/unadj.dupcor.sva.voomaw/meanvar-trends-PAGE2-CROP-RASTER.png
  7892. lyxscale 15
  7893. width 30col%
  7894. groupId voomaw-subfig
  7895. \end_inset
  7896. \end_layout
  7897. \begin_layout Plain Layout
  7898. \begin_inset Caption Standard
  7899. \begin_layout Plain Layout
  7900. \begin_inset CommandInset label
  7901. LatexCommand label
  7902. name "fig:meanvar-sva-voomaw"
  7903. \end_inset
  7904. Mean-variance trend after voom modeling in analysis C.
  7905. \end_layout
  7906. \end_inset
  7907. \end_layout
  7908. \end_inset
  7909. \end_layout
  7910. \begin_layout Plain Layout
  7911. \begin_inset Caption Standard
  7912. \begin_layout Plain Layout
  7913. \series bold
  7914. Mean-variance trend modeling in methylation array data.
  7915. \series default
  7916. The estimated log2(standard deviation) for each probe is plotted against
  7917. the probe's average M-value across all samples as a black point, with some
  7918. transparency to make overplotting more visible, since there are about 450,000
  7919. points.
  7920. Density of points is also indicated by the dark blue contour lines.
  7921. The prior variance trend estimated by eBayes is shown in light blue, while
  7922. the lowess trend of the points is shown in red.
  7923. \end_layout
  7924. \end_inset
  7925. \end_layout
  7926. \end_inset
  7927. \end_layout
  7928. \begin_layout Standard
  7929. \begin_inset ERT
  7930. status open
  7931. \begin_layout Plain Layout
  7932. \backslash
  7933. end{landscape}
  7934. \end_layout
  7935. \begin_layout Plain Layout
  7936. }
  7937. \end_layout
  7938. \end_inset
  7939. \end_layout
  7940. \begin_layout Standard
  7941. Figure
  7942. \begin_inset CommandInset ref
  7943. LatexCommand ref
  7944. reference "fig:meanvar-basic"
  7945. plural "false"
  7946. caps "false"
  7947. noprefix "false"
  7948. \end_inset
  7949. shows the relationship between the mean M-value and the standard deviation
  7950. calculated for each probe in the methylation array data set.
  7951. A few features of the data are apparent.
  7952. First, the data are very strongly bimodal, with peaks in the density around
  7953. M-values of +4 and -4.
  7954. These modes correspond to methylation sites that are nearly 100% methylated
  7955. and nearly 100% unmethylated, respectively.
  7956. The strong bomodality indicates that a majority of probes interrogate sites
  7957. that fall into one of these two categories.
  7958. The points in between these modes represent sites that are either partially
  7959. methylated in many samples, or are fully methylated in some samples and
  7960. fully unmethylated in other samples, or some combination.
  7961. The next visible feature of the data is the W-shaped variance trend.
  7962. The upticks in the variance trend on either side are expected, based on
  7963. the sigmoid transformation exaggerating small differences at extreme M-values
  7964. (Figure
  7965. \begin_inset CommandInset ref
  7966. LatexCommand ref
  7967. reference "fig:Sigmoid-beta-m-mapping"
  7968. plural "false"
  7969. caps "false"
  7970. noprefix "false"
  7971. \end_inset
  7972. ).
  7973. However, the uptick in the center is interesting: it indicates that sites
  7974. that are not constitutitively methylated or unmethylated have a higher
  7975. variance.
  7976. This could be a genuine biological effect, or it could be spurious noise
  7977. that is only observable at sites with varying methylation.
  7978. \end_layout
  7979. \begin_layout Standard
  7980. In Figure
  7981. \begin_inset CommandInset ref
  7982. LatexCommand ref
  7983. reference "fig:meanvar-sva-aw"
  7984. plural "false"
  7985. caps "false"
  7986. noprefix "false"
  7987. \end_inset
  7988. , we see the mean-variance trend for the same methylation array data, this
  7989. time with surrogate variables and sample quality weights estimated from
  7990. the data and included in the model.
  7991. As expected, the overall average variance is smaller, since the surrogate
  7992. variables account for some of the variance.
  7993. In addition, the uptick in variance in the middle of the M-value range
  7994. has disappeared, turning the W shape into a wide U shape.
  7995. This indicates that the excess variance in the probes with intermediate
  7996. M-values was explained by systematic variations not correlated with known
  7997. covariates, and these variations were modeled by the surrogate variables.
  7998. The result is a nearly flat variance trend for the entire intermediate
  7999. M-value range from about -3 to +3.
  8000. Note that this corresponds closely to the range within which the M-value
  8001. transformation shown in Figure
  8002. \begin_inset CommandInset ref
  8003. LatexCommand ref
  8004. reference "fig:Sigmoid-beta-m-mapping"
  8005. plural "false"
  8006. caps "false"
  8007. noprefix "false"
  8008. \end_inset
  8009. is nearly linear.
  8010. In contrast, the excess variance at the extremes (greater than +3 and less
  8011. than -3) was not
  8012. \begin_inset Quotes eld
  8013. \end_inset
  8014. absorbed
  8015. \begin_inset Quotes erd
  8016. \end_inset
  8017. by the surrogate variables and remains in the plot, indicating that this
  8018. variation has no systematic component: probes with extreme M-values are
  8019. uniformly more variable across all samples, as expected.
  8020. \end_layout
  8021. \begin_layout Standard
  8022. Figure
  8023. \begin_inset CommandInset ref
  8024. LatexCommand ref
  8025. reference "fig:meanvar-sva-voomaw"
  8026. plural "false"
  8027. caps "false"
  8028. noprefix "false"
  8029. \end_inset
  8030. shows the mean-variance trend after fitting the model with the observation
  8031. weights assigned by voom based on the mean-variance trend shown in Figure
  8032. \begin_inset CommandInset ref
  8033. LatexCommand ref
  8034. reference "fig:meanvar-sva-aw"
  8035. plural "false"
  8036. caps "false"
  8037. noprefix "false"
  8038. \end_inset
  8039. .
  8040. As expected, the weights exactly counteract the trend in the data, resulting
  8041. in a nearly flat trend centered vertically at 1 (i.e.
  8042. 0 on the log scale).
  8043. This shows that the observations with extreme M-values have been appropriately
  8044. down-weighted to account for the fact that the noise in those observations
  8045. has been amplified by the non-linear M-value transformation.
  8046. In turn, this gives relatively more weight to observervations in the middle
  8047. region, which are more likely to correspond to probes measuring interesting
  8048. biology (not constitutively methylated or unmethylated).
  8049. \end_layout
  8050. \begin_layout Standard
  8051. \begin_inset Float table
  8052. wide false
  8053. sideways false
  8054. status open
  8055. \begin_layout Plain Layout
  8056. \align center
  8057. \begin_inset Tabular
  8058. <lyxtabular version="3" rows="5" columns="3">
  8059. <features tabularvalignment="middle">
  8060. <column alignment="center" valignment="top">
  8061. <column alignment="center" valignment="top">
  8062. <column alignment="center" valignment="top">
  8063. <row>
  8064. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8065. \begin_inset Text
  8066. \begin_layout Plain Layout
  8067. Covariate
  8068. \end_layout
  8069. \end_inset
  8070. </cell>
  8071. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8072. \begin_inset Text
  8073. \begin_layout Plain Layout
  8074. Test used
  8075. \end_layout
  8076. \end_inset
  8077. </cell>
  8078. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  8079. \begin_inset Text
  8080. \begin_layout Plain Layout
  8081. p-value
  8082. \end_layout
  8083. \end_inset
  8084. </cell>
  8085. </row>
  8086. <row>
  8087. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8088. \begin_inset Text
  8089. \begin_layout Plain Layout
  8090. Transplant Status
  8091. \end_layout
  8092. \end_inset
  8093. </cell>
  8094. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8095. \begin_inset Text
  8096. \begin_layout Plain Layout
  8097. F-test
  8098. \end_layout
  8099. \end_inset
  8100. </cell>
  8101. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8102. \begin_inset Text
  8103. \begin_layout Plain Layout
  8104. 0.404
  8105. \end_layout
  8106. \end_inset
  8107. </cell>
  8108. </row>
  8109. <row>
  8110. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8111. \begin_inset Text
  8112. \begin_layout Plain Layout
  8113. Diabetes Diagnosis
  8114. \end_layout
  8115. \end_inset
  8116. </cell>
  8117. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8118. \begin_inset Text
  8119. \begin_layout Plain Layout
  8120. \emph on
  8121. t
  8122. \emph default
  8123. -test
  8124. \end_layout
  8125. \end_inset
  8126. </cell>
  8127. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8128. \begin_inset Text
  8129. \begin_layout Plain Layout
  8130. 0.00106
  8131. \end_layout
  8132. \end_inset
  8133. </cell>
  8134. </row>
  8135. <row>
  8136. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8137. \begin_inset Text
  8138. \begin_layout Plain Layout
  8139. Sex
  8140. \end_layout
  8141. \end_inset
  8142. </cell>
  8143. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8144. \begin_inset Text
  8145. \begin_layout Plain Layout
  8146. \emph on
  8147. t
  8148. \emph default
  8149. -test
  8150. \end_layout
  8151. \end_inset
  8152. </cell>
  8153. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8154. \begin_inset Text
  8155. \begin_layout Plain Layout
  8156. 0.148
  8157. \end_layout
  8158. \end_inset
  8159. </cell>
  8160. </row>
  8161. <row>
  8162. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8163. \begin_inset Text
  8164. \begin_layout Plain Layout
  8165. Age
  8166. \end_layout
  8167. \end_inset
  8168. </cell>
  8169. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8170. \begin_inset Text
  8171. \begin_layout Plain Layout
  8172. linear regression
  8173. \end_layout
  8174. \end_inset
  8175. </cell>
  8176. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  8177. \begin_inset Text
  8178. \begin_layout Plain Layout
  8179. 0.212
  8180. \end_layout
  8181. \end_inset
  8182. </cell>
  8183. </row>
  8184. </lyxtabular>
  8185. \end_inset
  8186. \end_layout
  8187. \begin_layout Plain Layout
  8188. \begin_inset Caption Standard
  8189. \begin_layout Plain Layout
  8190. \series bold
  8191. \begin_inset CommandInset label
  8192. LatexCommand label
  8193. name "tab:weight-covariate-tests"
  8194. \end_inset
  8195. Association of sample weights with clinical covariates in methylation array
  8196. data.
  8197. \series default
  8198. Computed sample quality log weights were tested for significant association
  8199. with each of the variables in the model (1st column).
  8200. An appropriate test was selected for each variable based on whether the
  8201. variable had 2 categories (
  8202. \emph on
  8203. t
  8204. \emph default
  8205. -test), had more than 2 categories (F-test), or was numeric (linear regression).
  8206. The test selected is shown in the 2nd column.
  8207. P-values for association with the log weights are shown in the 3rd column.
  8208. No multiple testing adjustment was performed for these p-values.
  8209. \end_layout
  8210. \end_inset
  8211. \end_layout
  8212. \end_inset
  8213. \end_layout
  8214. \begin_layout Standard
  8215. \begin_inset Float figure
  8216. wide false
  8217. sideways false
  8218. status open
  8219. \begin_layout Plain Layout
  8220. \begin_inset Flex TODO Note (inline)
  8221. status open
  8222. \begin_layout Plain Layout
  8223. Redo the sample weight boxplot with notches, and remove fill colors
  8224. \end_layout
  8225. \end_inset
  8226. \end_layout
  8227. \begin_layout Plain Layout
  8228. \align center
  8229. \begin_inset Graphics
  8230. filename graphics/methylvoom/unadj.dupcor.sva.voomaw/sample-weights-PAGE3-CROP.pdf
  8231. lyxscale 50
  8232. width 60col%
  8233. groupId colwidth
  8234. \end_inset
  8235. \end_layout
  8236. \begin_layout Plain Layout
  8237. \begin_inset Caption Standard
  8238. \begin_layout Plain Layout
  8239. \begin_inset CommandInset label
  8240. LatexCommand label
  8241. name "fig:diabetes-sample-weights"
  8242. \end_inset
  8243. \series bold
  8244. Box-and-whiskers plot of sample quality weights grouped by diabetes diagnosis.
  8245. \series default
  8246. Samples were grouped based on diabetes diagnosis, and the distribution of
  8247. sample quality weights for each diagnosis was plotted as a box-and-whiskers
  8248. plot
  8249. \begin_inset CommandInset citation
  8250. LatexCommand cite
  8251. key "McGill1978"
  8252. literal "false"
  8253. \end_inset
  8254. .
  8255. \end_layout
  8256. \end_inset
  8257. \end_layout
  8258. \begin_layout Plain Layout
  8259. \end_layout
  8260. \end_inset
  8261. \end_layout
  8262. \begin_layout Standard
  8263. To determine whether any of the known experimental factors had an impact
  8264. on data quality, the sample quality weights estimated from the data were
  8265. tested for association with each of the experimental factors (Table
  8266. \begin_inset CommandInset ref
  8267. LatexCommand ref
  8268. reference "tab:weight-covariate-tests"
  8269. plural "false"
  8270. caps "false"
  8271. noprefix "false"
  8272. \end_inset
  8273. ).
  8274. Diabetes diagnosis was found to have a potentially significant association
  8275. with the sample weights, with a t-test p-value of
  8276. \begin_inset Formula $1.06\times10^{-3}$
  8277. \end_inset
  8278. .
  8279. Figure
  8280. \begin_inset CommandInset ref
  8281. LatexCommand ref
  8282. reference "fig:diabetes-sample-weights"
  8283. plural "false"
  8284. caps "false"
  8285. noprefix "false"
  8286. \end_inset
  8287. shows the distribution of sample weights grouped by diabetes diagnosis.
  8288. The samples from patients with Type 2 diabetes were assigned significantly
  8289. lower weights than those from patients with Type 1 diabetes.
  8290. This indicates that the type 2 diabetes samples had an overall higher variance
  8291. on average across all probes.
  8292. \end_layout
  8293. \begin_layout Standard
  8294. \begin_inset Float table
  8295. wide false
  8296. sideways false
  8297. status open
  8298. \begin_layout Plain Layout
  8299. \align center
  8300. \begin_inset Flex TODO Note (inline)
  8301. status open
  8302. \begin_layout Plain Layout
  8303. Consider transposing these tables
  8304. \end_layout
  8305. \end_inset
  8306. \end_layout
  8307. \begin_layout Plain Layout
  8308. \begin_inset Float table
  8309. wide false
  8310. sideways false
  8311. status open
  8312. \begin_layout Plain Layout
  8313. \align center
  8314. \begin_inset Tabular
  8315. <lyxtabular version="3" rows="5" columns="4">
  8316. <features tabularvalignment="middle">
  8317. <column alignment="center" valignment="top">
  8318. <column alignment="center" valignment="top">
  8319. <column alignment="center" valignment="top">
  8320. <column alignment="center" valignment="top">
  8321. <row>
  8322. <cell alignment="center" valignment="top" usebox="none">
  8323. \begin_inset Text
  8324. \begin_layout Plain Layout
  8325. \end_layout
  8326. \end_inset
  8327. </cell>
  8328. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8329. \begin_inset Text
  8330. \begin_layout Plain Layout
  8331. Analysis
  8332. \end_layout
  8333. \end_inset
  8334. </cell>
  8335. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8336. \begin_inset Text
  8337. \begin_layout Plain Layout
  8338. \end_layout
  8339. \end_inset
  8340. </cell>
  8341. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8342. \begin_inset Text
  8343. \begin_layout Plain Layout
  8344. \end_layout
  8345. \end_inset
  8346. </cell>
  8347. </row>
  8348. <row>
  8349. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8350. \begin_inset Text
  8351. \begin_layout Plain Layout
  8352. Contrast
  8353. \end_layout
  8354. \end_inset
  8355. </cell>
  8356. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8357. \begin_inset Text
  8358. \begin_layout Plain Layout
  8359. A
  8360. \end_layout
  8361. \end_inset
  8362. </cell>
  8363. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8364. \begin_inset Text
  8365. \begin_layout Plain Layout
  8366. B
  8367. \end_layout
  8368. \end_inset
  8369. </cell>
  8370. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  8371. \begin_inset Text
  8372. \begin_layout Plain Layout
  8373. C
  8374. \end_layout
  8375. \end_inset
  8376. </cell>
  8377. </row>
  8378. <row>
  8379. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8380. \begin_inset Text
  8381. \begin_layout Plain Layout
  8382. TX vs AR
  8383. \end_layout
  8384. \end_inset
  8385. </cell>
  8386. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8387. \begin_inset Text
  8388. \begin_layout Plain Layout
  8389. 0
  8390. \end_layout
  8391. \end_inset
  8392. </cell>
  8393. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8394. \begin_inset Text
  8395. \begin_layout Plain Layout
  8396. 25
  8397. \end_layout
  8398. \end_inset
  8399. </cell>
  8400. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8401. \begin_inset Text
  8402. \begin_layout Plain Layout
  8403. 22
  8404. \end_layout
  8405. \end_inset
  8406. </cell>
  8407. </row>
  8408. <row>
  8409. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8410. \begin_inset Text
  8411. \begin_layout Plain Layout
  8412. TX vs ADNR
  8413. \end_layout
  8414. \end_inset
  8415. </cell>
  8416. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8417. \begin_inset Text
  8418. \begin_layout Plain Layout
  8419. 7
  8420. \end_layout
  8421. \end_inset
  8422. </cell>
  8423. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8424. \begin_inset Text
  8425. \begin_layout Plain Layout
  8426. 338
  8427. \end_layout
  8428. \end_inset
  8429. </cell>
  8430. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8431. \begin_inset Text
  8432. \begin_layout Plain Layout
  8433. 369
  8434. \end_layout
  8435. \end_inset
  8436. </cell>
  8437. </row>
  8438. <row>
  8439. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8440. \begin_inset Text
  8441. \begin_layout Plain Layout
  8442. TX vs CAN
  8443. \end_layout
  8444. \end_inset
  8445. </cell>
  8446. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8447. \begin_inset Text
  8448. \begin_layout Plain Layout
  8449. 0
  8450. \end_layout
  8451. \end_inset
  8452. </cell>
  8453. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8454. \begin_inset Text
  8455. \begin_layout Plain Layout
  8456. 231
  8457. \end_layout
  8458. \end_inset
  8459. </cell>
  8460. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  8461. \begin_inset Text
  8462. \begin_layout Plain Layout
  8463. 278
  8464. \end_layout
  8465. \end_inset
  8466. </cell>
  8467. </row>
  8468. </lyxtabular>
  8469. \end_inset
  8470. \end_layout
  8471. \begin_layout Plain Layout
  8472. \begin_inset Caption Standard
  8473. \begin_layout Plain Layout
  8474. \begin_inset CommandInset label
  8475. LatexCommand label
  8476. name "tab:methyl-num-signif"
  8477. \end_inset
  8478. Number of probes significant at 10% FDR.
  8479. \end_layout
  8480. \end_inset
  8481. \end_layout
  8482. \end_inset
  8483. \begin_inset space \hfill{}
  8484. \end_inset
  8485. \begin_inset Float table
  8486. wide false
  8487. sideways false
  8488. status open
  8489. \begin_layout Plain Layout
  8490. \align center
  8491. \begin_inset Tabular
  8492. <lyxtabular version="3" rows="5" columns="4">
  8493. <features tabularvalignment="middle">
  8494. <column alignment="center" valignment="top">
  8495. <column alignment="center" valignment="top">
  8496. <column alignment="center" valignment="top">
  8497. <column alignment="center" valignment="top">
  8498. <row>
  8499. <cell alignment="center" valignment="top" usebox="none">
  8500. \begin_inset Text
  8501. \begin_layout Plain Layout
  8502. \end_layout
  8503. \end_inset
  8504. </cell>
  8505. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8506. \begin_inset Text
  8507. \begin_layout Plain Layout
  8508. Analysis
  8509. \end_layout
  8510. \end_inset
  8511. </cell>
  8512. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8513. \begin_inset Text
  8514. \begin_layout Plain Layout
  8515. \end_layout
  8516. \end_inset
  8517. </cell>
  8518. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8519. \begin_inset Text
  8520. \begin_layout Plain Layout
  8521. \end_layout
  8522. \end_inset
  8523. </cell>
  8524. </row>
  8525. <row>
  8526. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8527. \begin_inset Text
  8528. \begin_layout Plain Layout
  8529. Contrast
  8530. \end_layout
  8531. \end_inset
  8532. </cell>
  8533. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8534. \begin_inset Text
  8535. \begin_layout Plain Layout
  8536. A
  8537. \end_layout
  8538. \end_inset
  8539. </cell>
  8540. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8541. \begin_inset Text
  8542. \begin_layout Plain Layout
  8543. B
  8544. \end_layout
  8545. \end_inset
  8546. </cell>
  8547. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  8548. \begin_inset Text
  8549. \begin_layout Plain Layout
  8550. C
  8551. \end_layout
  8552. \end_inset
  8553. </cell>
  8554. </row>
  8555. <row>
  8556. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8557. \begin_inset Text
  8558. \begin_layout Plain Layout
  8559. TX vs AR
  8560. \end_layout
  8561. \end_inset
  8562. </cell>
  8563. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8564. \begin_inset Text
  8565. \begin_layout Plain Layout
  8566. 0
  8567. \end_layout
  8568. \end_inset
  8569. </cell>
  8570. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8571. \begin_inset Text
  8572. \begin_layout Plain Layout
  8573. 10,063
  8574. \end_layout
  8575. \end_inset
  8576. </cell>
  8577. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8578. \begin_inset Text
  8579. \begin_layout Plain Layout
  8580. 11,225
  8581. \end_layout
  8582. \end_inset
  8583. </cell>
  8584. </row>
  8585. <row>
  8586. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8587. \begin_inset Text
  8588. \begin_layout Plain Layout
  8589. TX vs ADNR
  8590. \end_layout
  8591. \end_inset
  8592. </cell>
  8593. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8594. \begin_inset Text
  8595. \begin_layout Plain Layout
  8596. 27
  8597. \end_layout
  8598. \end_inset
  8599. </cell>
  8600. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  8601. \begin_inset Text
  8602. \begin_layout Plain Layout
  8603. 12,674
  8604. \end_layout
  8605. \end_inset
  8606. </cell>
  8607. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  8608. \begin_inset Text
  8609. \begin_layout Plain Layout
  8610. 13,086
  8611. \end_layout
  8612. \end_inset
  8613. </cell>
  8614. </row>
  8615. <row>
  8616. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8617. \begin_inset Text
  8618. \begin_layout Plain Layout
  8619. TX vs CAN
  8620. \end_layout
  8621. \end_inset
  8622. </cell>
  8623. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8624. \begin_inset Text
  8625. \begin_layout Plain Layout
  8626. 966
  8627. \end_layout
  8628. \end_inset
  8629. </cell>
  8630. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  8631. \begin_inset Text
  8632. \begin_layout Plain Layout
  8633. 20,039
  8634. \end_layout
  8635. \end_inset
  8636. </cell>
  8637. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  8638. \begin_inset Text
  8639. \begin_layout Plain Layout
  8640. 20,955
  8641. \end_layout
  8642. \end_inset
  8643. </cell>
  8644. </row>
  8645. </lyxtabular>
  8646. \end_inset
  8647. \end_layout
  8648. \begin_layout Plain Layout
  8649. \begin_inset Caption Standard
  8650. \begin_layout Plain Layout
  8651. \begin_inset CommandInset label
  8652. LatexCommand label
  8653. name "tab:methyl-est-nonnull"
  8654. \end_inset
  8655. Estimated number of non-null tests, using the method of averaging local
  8656. FDR values
  8657. \begin_inset CommandInset citation
  8658. LatexCommand cite
  8659. key "Phipson2013Thesis"
  8660. literal "false"
  8661. \end_inset
  8662. .
  8663. \end_layout
  8664. \end_inset
  8665. \end_layout
  8666. \end_inset
  8667. \end_layout
  8668. \begin_layout Plain Layout
  8669. \begin_inset Caption Standard
  8670. \begin_layout Plain Layout
  8671. \series bold
  8672. Estimates of degree of differential methylation in for each contrast in
  8673. each analysis.
  8674. \series default
  8675. For each of the analyses in Table
  8676. \begin_inset CommandInset ref
  8677. LatexCommand ref
  8678. reference "tab:Summary-of-meth-analysis"
  8679. plural "false"
  8680. caps "false"
  8681. noprefix "false"
  8682. \end_inset
  8683. , these tables show the number of probes called significantly differentially
  8684. methylated at a threshold of 10% FDR for each comparison between TX and
  8685. the other 3 transplant statuses (a) and the estimated total number of probes
  8686. that are differentially methylated (b).
  8687. \end_layout
  8688. \end_inset
  8689. \end_layout
  8690. \end_inset
  8691. \end_layout
  8692. \begin_layout Standard
  8693. \begin_inset Float figure
  8694. wide false
  8695. sideways false
  8696. status open
  8697. \begin_layout Plain Layout
  8698. \align center
  8699. \series bold
  8700. \begin_inset Float figure
  8701. wide false
  8702. sideways false
  8703. status collapsed
  8704. \begin_layout Plain Layout
  8705. \align center
  8706. \begin_inset Graphics
  8707. filename graphics/methylvoom/unadj.dupcor/pval-histograms-PAGE1.pdf
  8708. lyxscale 33
  8709. width 30col%
  8710. groupId meth-pval-hist
  8711. \end_inset
  8712. \end_layout
  8713. \begin_layout Plain Layout
  8714. \series bold
  8715. \begin_inset Caption Standard
  8716. \begin_layout Plain Layout
  8717. AR vs.
  8718. TX, Analysis A
  8719. \end_layout
  8720. \end_inset
  8721. \end_layout
  8722. \begin_layout Plain Layout
  8723. \end_layout
  8724. \end_inset
  8725. \begin_inset space \hfill{}
  8726. \end_inset
  8727. \begin_inset Float figure
  8728. wide false
  8729. sideways false
  8730. status collapsed
  8731. \begin_layout Plain Layout
  8732. \align center
  8733. \begin_inset Graphics
  8734. filename graphics/methylvoom/unadj.dupcor/pval-histograms-PAGE2.pdf
  8735. lyxscale 33
  8736. width 30col%
  8737. groupId meth-pval-hist
  8738. \end_inset
  8739. \end_layout
  8740. \begin_layout Plain Layout
  8741. \series bold
  8742. \begin_inset Caption Standard
  8743. \begin_layout Plain Layout
  8744. ADNR vs.
  8745. TX, Analysis A
  8746. \end_layout
  8747. \end_inset
  8748. \end_layout
  8749. \end_inset
  8750. \begin_inset space \hfill{}
  8751. \end_inset
  8752. \begin_inset Float figure
  8753. wide false
  8754. sideways false
  8755. status collapsed
  8756. \begin_layout Plain Layout
  8757. \align center
  8758. \begin_inset Graphics
  8759. filename graphics/methylvoom/unadj.dupcor/pval-histograms-PAGE3.pdf
  8760. lyxscale 33
  8761. width 30col%
  8762. groupId meth-pval-hist
  8763. \end_inset
  8764. \end_layout
  8765. \begin_layout Plain Layout
  8766. \series bold
  8767. \begin_inset Caption Standard
  8768. \begin_layout Plain Layout
  8769. CAN vs.
  8770. TX, Analysis A
  8771. \end_layout
  8772. \end_inset
  8773. \end_layout
  8774. \end_inset
  8775. \end_layout
  8776. \begin_layout Plain Layout
  8777. \align center
  8778. \series bold
  8779. \begin_inset Float figure
  8780. wide false
  8781. sideways false
  8782. status collapsed
  8783. \begin_layout Plain Layout
  8784. \align center
  8785. \begin_inset Graphics
  8786. filename graphics/methylvoom/unadj.dupcor.sva.aw/pval-histograms-PAGE1.pdf
  8787. lyxscale 33
  8788. width 30col%
  8789. groupId meth-pval-hist
  8790. \end_inset
  8791. \end_layout
  8792. \begin_layout Plain Layout
  8793. \series bold
  8794. \begin_inset Caption Standard
  8795. \begin_layout Plain Layout
  8796. AR vs.
  8797. TX, Analysis B
  8798. \end_layout
  8799. \end_inset
  8800. \end_layout
  8801. \end_inset
  8802. \begin_inset space \hfill{}
  8803. \end_inset
  8804. \begin_inset Float figure
  8805. wide false
  8806. sideways false
  8807. status collapsed
  8808. \begin_layout Plain Layout
  8809. \align center
  8810. \begin_inset Graphics
  8811. filename graphics/methylvoom/unadj.dupcor.sva.aw/pval-histograms-PAGE2.pdf
  8812. lyxscale 33
  8813. width 30col%
  8814. groupId meth-pval-hist
  8815. \end_inset
  8816. \end_layout
  8817. \begin_layout Plain Layout
  8818. \series bold
  8819. \begin_inset Caption Standard
  8820. \begin_layout Plain Layout
  8821. ADNR vs.
  8822. TX, Analysis B
  8823. \end_layout
  8824. \end_inset
  8825. \end_layout
  8826. \end_inset
  8827. \begin_inset space \hfill{}
  8828. \end_inset
  8829. \begin_inset Float figure
  8830. wide false
  8831. sideways false
  8832. status collapsed
  8833. \begin_layout Plain Layout
  8834. \align center
  8835. \begin_inset Graphics
  8836. filename graphics/methylvoom/unadj.dupcor.sva.aw/pval-histograms-PAGE3.pdf
  8837. lyxscale 33
  8838. width 30col%
  8839. groupId meth-pval-hist
  8840. \end_inset
  8841. \end_layout
  8842. \begin_layout Plain Layout
  8843. \series bold
  8844. \begin_inset Caption Standard
  8845. \begin_layout Plain Layout
  8846. CAN vs.
  8847. TX, Analysis B
  8848. \end_layout
  8849. \end_inset
  8850. \end_layout
  8851. \end_inset
  8852. \end_layout
  8853. \begin_layout Plain Layout
  8854. \align center
  8855. \series bold
  8856. \begin_inset Float figure
  8857. wide false
  8858. sideways false
  8859. status collapsed
  8860. \begin_layout Plain Layout
  8861. \align center
  8862. \begin_inset Graphics
  8863. filename graphics/methylvoom/unadj.dupcor.sva.voomaw/pval-histograms-PAGE1.pdf
  8864. lyxscale 33
  8865. width 30col%
  8866. groupId meth-pval-hist
  8867. \end_inset
  8868. \end_layout
  8869. \begin_layout Plain Layout
  8870. \series bold
  8871. \begin_inset Caption Standard
  8872. \begin_layout Plain Layout
  8873. AR vs.
  8874. TX, Analysis C
  8875. \end_layout
  8876. \end_inset
  8877. \end_layout
  8878. \end_inset
  8879. \begin_inset space \hfill{}
  8880. \end_inset
  8881. \begin_inset Float figure
  8882. wide false
  8883. sideways false
  8884. status collapsed
  8885. \begin_layout Plain Layout
  8886. \align center
  8887. \begin_inset Graphics
  8888. filename graphics/methylvoom/unadj.dupcor.sva.voomaw/pval-histograms-PAGE2.pdf
  8889. lyxscale 33
  8890. width 30col%
  8891. groupId meth-pval-hist
  8892. \end_inset
  8893. \end_layout
  8894. \begin_layout Plain Layout
  8895. \series bold
  8896. \begin_inset Caption Standard
  8897. \begin_layout Plain Layout
  8898. ADNR vs.
  8899. TX, Analysis C
  8900. \end_layout
  8901. \end_inset
  8902. \end_layout
  8903. \end_inset
  8904. \begin_inset space \hfill{}
  8905. \end_inset
  8906. \begin_inset Float figure
  8907. wide false
  8908. sideways false
  8909. status collapsed
  8910. \begin_layout Plain Layout
  8911. \align center
  8912. \begin_inset Graphics
  8913. filename graphics/methylvoom/unadj.dupcor.sva.voomaw/pval-histograms-PAGE3.pdf
  8914. lyxscale 33
  8915. width 30col%
  8916. groupId meth-pval-hist
  8917. \end_inset
  8918. \end_layout
  8919. \begin_layout Plain Layout
  8920. \series bold
  8921. \begin_inset Caption Standard
  8922. \begin_layout Plain Layout
  8923. CAN vs.
  8924. TX, Analysis C
  8925. \end_layout
  8926. \end_inset
  8927. \end_layout
  8928. \end_inset
  8929. \end_layout
  8930. \begin_layout Plain Layout
  8931. \begin_inset Caption Standard
  8932. \begin_layout Plain Layout
  8933. \series bold
  8934. \begin_inset CommandInset label
  8935. LatexCommand label
  8936. name "fig:meth-p-value-histograms"
  8937. \end_inset
  8938. Probe p-value histograms for each contrast in each analysis.
  8939. \series default
  8940. For each differential methylation test of interest, the distribution of
  8941. p-values across all probes is plotted as a histogram.
  8942. The red solid line indicates the density that would be expected under the
  8943. null hypothesis for all probes (a
  8944. \begin_inset Formula $\mathrm{Uniform}(0,1)$
  8945. \end_inset
  8946. distribution), while the blue dotted line indicates the fraction of p-values
  8947. that actually follow the null hypothesis (
  8948. \begin_inset Formula $\hat{\pi}_{0}$
  8949. \end_inset
  8950. ) estimated using the method of averaging local FDR values
  8951. \begin_inset CommandInset citation
  8952. LatexCommand cite
  8953. key "Phipson2013Thesis"
  8954. literal "false"
  8955. \end_inset
  8956. .
  8957. the blue line is only shown in each plot if the estimate of
  8958. \begin_inset Formula $\hat{\pi}_{0}$
  8959. \end_inset
  8960. for that p-value distribution is different from 1.
  8961. \end_layout
  8962. \end_inset
  8963. \end_layout
  8964. \end_inset
  8965. \end_layout
  8966. \begin_layout Standard
  8967. Table
  8968. \begin_inset CommandInset ref
  8969. LatexCommand ref
  8970. reference "tab:methyl-num-signif"
  8971. plural "false"
  8972. caps "false"
  8973. noprefix "false"
  8974. \end_inset
  8975. shows the number of significantly differentially methylated probes reported
  8976. by each analysis for each comparison of interest at an FDR of 10%.
  8977. As expected, the more elaborate analyses, B and C, report more significant
  8978. probes than the more basic analysis A, consistent with the conclusions
  8979. above that the data contain hidden systematic variations that must be modeled.
  8980. Table
  8981. \begin_inset CommandInset ref
  8982. LatexCommand ref
  8983. reference "tab:methyl-est-nonnull"
  8984. plural "false"
  8985. caps "false"
  8986. noprefix "false"
  8987. \end_inset
  8988. shows the estimated number differentially methylated probes for each test
  8989. from each analysis.
  8990. This was computed by estimating the proportion of null hypotheses that
  8991. were true using the method of
  8992. \begin_inset CommandInset citation
  8993. LatexCommand cite
  8994. key "Phipson2013Thesis"
  8995. literal "false"
  8996. \end_inset
  8997. and subtracting that fraction from the total number of probes, yielding
  8998. an estimate of the number of null hypotheses that are false based on the
  8999. distribution of p-values across the entire dataset.
  9000. Note that this does not identify which null hypotheses should be rejected
  9001. (i.e.
  9002. which probes are significant); it only estimates the true number of such
  9003. probes.
  9004. Once again, analyses B and C result it much larger estimates for the number
  9005. of differentially methylated probes.
  9006. In this case, analysis C, the only analysis that includes voom, estimates
  9007. the largest number of differentially methylated probes for all 3 contrasts.
  9008. If the assumptions of all the methods employed hold, then this represents
  9009. a gain in statistical power over the simpler analysis A.
  9010. Figure
  9011. \begin_inset CommandInset ref
  9012. LatexCommand ref
  9013. reference "fig:meth-p-value-histograms"
  9014. plural "false"
  9015. caps "false"
  9016. noprefix "false"
  9017. \end_inset
  9018. shows the p-value distributions for each test, from which the numbers in
  9019. Table
  9020. \begin_inset CommandInset ref
  9021. LatexCommand ref
  9022. reference "tab:methyl-est-nonnull"
  9023. plural "false"
  9024. caps "false"
  9025. noprefix "false"
  9026. \end_inset
  9027. were generated.
  9028. The distributions for analysis A all have a dip in density near zero, which
  9029. is a strong sign of a poor model fit.
  9030. The histograms for analyses B and C are more well-behaved, with a uniform
  9031. component stretching all the way from 0 to 1 representing the probes for
  9032. which the null hypotheses is true (no differential methylation), and a
  9033. zero-biased component representing the probes for which the null hypothesis
  9034. is false (differentially methylated).
  9035. These histograms do not indicate any major issues with the model fit.
  9036. \end_layout
  9037. \begin_layout Standard
  9038. \begin_inset Flex TODO Note (inline)
  9039. status open
  9040. \begin_layout Plain Layout
  9041. If time allows, maybe generate the PCA plots before/after SVA effect subtraction
  9042. ?
  9043. \end_layout
  9044. \end_inset
  9045. \end_layout
  9046. \begin_layout Section
  9047. Discussion
  9048. \end_layout
  9049. \begin_layout Subsection
  9050. fRMA achieves clinically applicable normalization without sacrificing classifica
  9051. tion performance
  9052. \end_layout
  9053. \begin_layout Standard
  9054. As shown in Figure
  9055. \begin_inset CommandInset ref
  9056. LatexCommand ref
  9057. reference "fig:Classifier-probabilities-RMA"
  9058. plural "false"
  9059. caps "false"
  9060. noprefix "false"
  9061. \end_inset
  9062. , improper normalization, particularly separate normalization of training
  9063. and test samples, leads to unwanted biases in classification.
  9064. In a controlled experimental context, it is always possible to correct
  9065. this issue by normalizing all experimental samples together.
  9066. However, because it is not feasible to normalize all samples together in
  9067. a clinical context, a single-channel normalization is required is required.
  9068. \end_layout
  9069. \begin_layout Standard
  9070. The major concern in using a single-channel normalization is that non-single-cha
  9071. nnel methods can share information between arrays to improve the normalization,
  9072. and single-channel methods risk sacrificing the gains in normalization
  9073. accuracy that come from this information sharing.
  9074. In the case of RMA, this information sharing is accomplished through quantile
  9075. normalization and median polish steps.
  9076. The need for information sharing in quantile normalization can easily be
  9077. removed by learning a fixed set of quantiles from external data and normalizing
  9078. each array to these fixed quantiles, instead of the quantiles of the data
  9079. itself.
  9080. As long as the fixed quantiles are reasonable, the result will be similar
  9081. to standard RMA.
  9082. However, there is no analogous way to eliminate cross-array information
  9083. sharing in the median polish step, so fRMA replaces this with a weighted
  9084. average of probes on each array, with the weights learned from external
  9085. data.
  9086. This step of fRMA has the greatest potential to diverge from RMA un undesirable
  9087. ways.
  9088. \end_layout
  9089. \begin_layout Standard
  9090. However, when run on real data, fRMA performed at least as well as RMA in
  9091. both the internal validation and external validation tests.
  9092. This shows that fRMA can be used to normalize individual clinical samples
  9093. in a class prediction context without sacrificing the classifier performance
  9094. that would be obtained by using the more well-established RMA for normalization.
  9095. The other single-channel normalization method considered, SCAN, showed
  9096. some loss of AUC in the external validation test.
  9097. Based on these results, fRMA is the preferred normalization for clinical
  9098. samples in a class prediction context.
  9099. \end_layout
  9100. \begin_layout Subsection
  9101. Robust fRMA vectors can be generated for new array platforms
  9102. \end_layout
  9103. \begin_layout Standard
  9104. \begin_inset Flex TODO Note (inline)
  9105. status open
  9106. \begin_layout Plain Layout
  9107. Look up the exact numbers, do a find & replace for
  9108. \begin_inset Quotes eld
  9109. \end_inset
  9110. 850
  9111. \begin_inset Quotes erd
  9112. \end_inset
  9113. \end_layout
  9114. \end_inset
  9115. \end_layout
  9116. \begin_layout Standard
  9117. The published fRMA normalization vectors for the hgu133plus2 platform were
  9118. generated from a set of about 850 samples chosen from a wide range of tissues,
  9119. which the authors determined was sufficient to generate a robust set of
  9120. normalization vectors that could be applied across all tissues
  9121. \begin_inset CommandInset citation
  9122. LatexCommand cite
  9123. key "McCall2010"
  9124. literal "false"
  9125. \end_inset
  9126. .
  9127. Since we only had hthgu133pluspm for 2 tissues of interest, our needs were
  9128. more modest.
  9129. Even using only 130 samples in 26 batches of 5 samples each for kidney
  9130. biopsies, we were able to train a robust set of fRMA normalization vectors
  9131. that were not meaningfully affected by the random selection of 5 samples
  9132. from each batch.
  9133. As expected, the training process was just as robust for the blood samples
  9134. with 230 samples in 46 batches of 5 samples each.
  9135. Because these vectors were each generated using training samples from a
  9136. single tissue, they are not suitable for general use, unlike the vectors
  9137. provided with fRMA itself.
  9138. They are purpose-built for normalizing a specific type of sample on a specific
  9139. platform.
  9140. This is a mostly acceptable limitation in the context of developing a machine
  9141. learning classifier for diagnosing a disease based on samples of a specific
  9142. tissue.
  9143. \end_layout
  9144. \begin_layout Standard
  9145. \begin_inset Flex TODO Note (inline)
  9146. status open
  9147. \begin_layout Plain Layout
  9148. Talk about how these vectors can be used for any data from these tissues
  9149. on this platform even though they were custom made for this data set.
  9150. \end_layout
  9151. \end_inset
  9152. \end_layout
  9153. \begin_layout Standard
  9154. \begin_inset Flex TODO Note (inline)
  9155. status open
  9156. \begin_layout Plain Layout
  9157. How to bring up that these custom vectors were used in another project by
  9158. someone else that was never published?
  9159. \end_layout
  9160. \end_inset
  9161. \end_layout
  9162. \begin_layout Subsection
  9163. Methylation array data can be successfully analyzed using existing techniques,
  9164. but machine learning poses additional challenges
  9165. \end_layout
  9166. \begin_layout Standard
  9167. Both analysis strategies B and C both yield a reasonable analysis, with
  9168. a mean-variance trend that matches the expected behavior for the non-linear
  9169. M-value transformation (Figure
  9170. \begin_inset CommandInset ref
  9171. LatexCommand ref
  9172. reference "fig:meanvar-sva-aw"
  9173. plural "false"
  9174. caps "false"
  9175. noprefix "false"
  9176. \end_inset
  9177. ) and well-behaved p-value distributions (Figure
  9178. \begin_inset CommandInset ref
  9179. LatexCommand ref
  9180. reference "fig:meth-p-value-histograms"
  9181. plural "false"
  9182. caps "false"
  9183. noprefix "false"
  9184. \end_inset
  9185. ).
  9186. These two analyses also yield similar numbers of significant probes (Table
  9187. \begin_inset CommandInset ref
  9188. LatexCommand ref
  9189. reference "tab:methyl-num-signif"
  9190. plural "false"
  9191. caps "false"
  9192. noprefix "false"
  9193. \end_inset
  9194. ) and similar estimates of the number of differentially methylated probes
  9195. (Table
  9196. \begin_inset CommandInset ref
  9197. LatexCommand ref
  9198. reference "tab:methyl-est-nonnull"
  9199. plural "false"
  9200. caps "false"
  9201. noprefix "false"
  9202. \end_inset
  9203. ).
  9204. The main difference between these two analyses is the method used to account
  9205. for the mean-variance trend.
  9206. In analysis B, the trend is estimated and applied at the probe level: each
  9207. probe's estimated variance is squeezed toward the trend using an empirical
  9208. Bayes procedure (Figure
  9209. \begin_inset CommandInset ref
  9210. LatexCommand ref
  9211. reference "fig:meanvar-sva-aw"
  9212. plural "false"
  9213. caps "false"
  9214. noprefix "false"
  9215. \end_inset
  9216. ).
  9217. In analysis C, the trend is still estimated at the probe level, but instead
  9218. of estimating a single variance value shared across all observations for
  9219. a given probe, the voom method computes an initial estiamte of the variance
  9220. for each observation individually based on where its model-fitted M-value
  9221. falls on the trend line and then assigns inverse-variance weights to model
  9222. the difference in variance between observations.
  9223. An overall variance is still estimated for each probe using the same empirical
  9224. Bayes method, but now the residual trend is flat (Figure
  9225. \begin_inset CommandInset ref
  9226. LatexCommand ref
  9227. reference "fig:meanvar-sva-voomaw"
  9228. plural "false"
  9229. caps "false"
  9230. noprefix "false"
  9231. \end_inset
  9232. ), indicating that the mean-variance trend is adequately modeled by scaling
  9233. the estimated variance for each observation using the weights computed
  9234. by voom.
  9235. \end_layout
  9236. \begin_layout Standard
  9237. The difference between the standard empirical Bayes trended variance modeling
  9238. (analysis B) and voom (analysis C) is analogous to the difference between
  9239. a t-test with equal variance and a t-test with unequal variance, except
  9240. that the unequal group variances used in the latter test are estimated
  9241. based on the mean-variance trend from all the probes rather than the data
  9242. for the specific probe being tested, thus stabilizing the group variance
  9243. estimates by sharing information between probes.
  9244. Allowing voom to model the variance using observation weights in this manner
  9245. allows the linear model fit to concentrate statistical power where it will
  9246. do the most good.
  9247. For example, if a particular probe's M-values are always at the extreme
  9248. of the M-value range (e.g.
  9249. less than -4) for ADNR samples, but the M-values for that probe in TX and
  9250. CAN samples are within the flat region of the mean-variance trend (between
  9251. -3 and +3), voom is able to down-weight the contribution of the high-variance
  9252. M-values from the ADNR samples in order to gain more statistical power
  9253. while testing for differential methylation between TX and CAN.
  9254. In contrast, modeling the mean-variance trend only at the probe level would
  9255. combine the high-variance ADNR samples and lower-variance samples from
  9256. other conditions and estimate an intermediate variance for this probe.
  9257. In practice, analysis B shows that this approach is adequate, but the voom
  9258. approach in analysis C is at least as good on all model fit criteria and
  9259. yields a larger estimate for the number of differentially methylated genes,
  9260. \emph on
  9261. and
  9262. \emph default
  9263. it matches up better with the theoretical
  9264. \end_layout
  9265. \begin_layout Standard
  9266. The significant association of diebetes diagnosis with sample quality is
  9267. interesting.
  9268. The samples with Type 2 diabetes tended to have more variation, averaged
  9269. across all probes, than those with Type 1 diabetes.
  9270. This is consistent with the consensus that type 2 disbetes and the associated
  9271. metabolic syndrome represent a broad dysregulation of the body's endocrine
  9272. signalling related to metabolism [citation needed].
  9273. This dysregulation could easily manifest as a greater degree of variation
  9274. in the DNA methylation patterns of affected tissues.
  9275. In contrast, Type 1 disbetes has a more specific cause and effect, so a
  9276. less variable methylation signature is expected.
  9277. \end_layout
  9278. \begin_layout Standard
  9279. This preliminary anlaysis suggests that some degree of differential methylation
  9280. exists between TX and each of the three types of transplant disfunction
  9281. studied.
  9282. Hence, it may be feasible to train a classifier to diagnose transplant
  9283. disfunction from DNA methylation array data.
  9284. However, the major importance of both SVA and sample quality weighting
  9285. for proper modeling of this data poses significant challenges for any attempt
  9286. at a machine learning on data of similar quality.
  9287. While these are easily used in a modeling context with full sample information,
  9288. neither of these methods is directly applicable in a machine learning context,
  9289. where the diagnosis is not known ahead of time.
  9290. If a machine learning approach for methylation-based diagnosis is to be
  9291. pursued, it will either require machine-learning-friendly methods to address
  9292. the same systematic trends in the data that SVA and sample quality weighting
  9293. address, or it will require higher quality data with substantially less
  9294. systematic perturbation of the data.
  9295. \end_layout
  9296. \begin_layout Section
  9297. Future Directions
  9298. \end_layout
  9299. \begin_layout Standard
  9300. \begin_inset Flex TODO Note (inline)
  9301. status open
  9302. \begin_layout Plain Layout
  9303. Some work was already being done with the existing fRMA vectors.
  9304. Do I mention that here?
  9305. \end_layout
  9306. \end_inset
  9307. \end_layout
  9308. \begin_layout Subsection
  9309. Improving fRMA to allow training from batches of unequal size
  9310. \end_layout
  9311. \begin_layout Standard
  9312. Because the tools for building fRMA normalization vectors require equal-size
  9313. batches, many samples must be discarded from the training data.
  9314. This is undesirable for a few reasons.
  9315. First, more data is simply better, all other things being equal.
  9316. In this case,
  9317. \begin_inset Quotes eld
  9318. \end_inset
  9319. better
  9320. \begin_inset Quotes erd
  9321. \end_inset
  9322. means a more precise estimate of normalization parameters.
  9323. In addition, the samples to be discarded must be chosen arbitrarily, which
  9324. introduces an unnecessary element of randomness into the estimation process.
  9325. While the randomness can be made deterministic by setting a consistent
  9326. random seed, the need for equal size batches also introduces a need for
  9327. the analyst to decide on the appropriate trade-off between batch size and
  9328. the number of batches.
  9329. This introduces an unnecessary and undesirable
  9330. \begin_inset Quotes eld
  9331. \end_inset
  9332. researcher degree of freedom
  9333. \begin_inset Quotes erd
  9334. \end_inset
  9335. into the analysis, since the generated normalization vectors now depend
  9336. on the choice of batch size based on vague selection criteria and instinct,
  9337. which can unintentionally inproduce bias if the researcher chooses a batch
  9338. size based on what seems to yield the most favorable downstream results
  9339. \begin_inset CommandInset citation
  9340. LatexCommand cite
  9341. key "Simmons2011"
  9342. literal "false"
  9343. \end_inset
  9344. .
  9345. \end_layout
  9346. \begin_layout Standard
  9347. Fortunately, the requirement for equal-size batches is not inherent to the
  9348. fRMA algorithm but rather a limitation of the implementation in the frmaTools
  9349. package.
  9350. In personal communication, the package's author, Matthew McCall, has indicated
  9351. that with some work, it should be possible to improve the implementation
  9352. to work with batches of unequal sizes.
  9353. The current implementation ignores the batch size when calculating with-batch
  9354. and between-batch residual variances, since the batch size constant cancels
  9355. out later in the calculations as long as all batches are of equal size.
  9356. Hence, the calculations of these parameters would need to be modified to
  9357. remove this optimization and properly calculate the variances using the
  9358. full formula.
  9359. Once this modification is made, a new strategy would need to be developed
  9360. for assessing the stability of parameter estimates, since the random subsamplin
  9361. g step is eliminated, meaning that different subsamplings can no longer
  9362. be compared as in Figures
  9363. \begin_inset CommandInset ref
  9364. LatexCommand ref
  9365. reference "fig:frma-violin"
  9366. plural "false"
  9367. caps "false"
  9368. noprefix "false"
  9369. \end_inset
  9370. and
  9371. \begin_inset CommandInset ref
  9372. LatexCommand ref
  9373. reference "fig:Representative-MA-plots"
  9374. plural "false"
  9375. caps "false"
  9376. noprefix "false"
  9377. \end_inset
  9378. .
  9379. Bootstrap resampling is likely a good candidate here: sample many training
  9380. sets of equal size from the existing training set with replacement, estimate
  9381. parameters from each resampled training set, and compare the estimated
  9382. parameters between bootstraps in order to quantify the variability in each
  9383. parameter's estimation.
  9384. \end_layout
  9385. \begin_layout Subsection
  9386. Developing methylation arrays as a diagnostic tool for kidney transplant
  9387. rejection
  9388. \end_layout
  9389. \begin_layout Standard
  9390. The current study has showed that DNA methylation, as assayed by Illumina
  9391. 450k methylation arrays, has some potential for diagnosing transplant dysfuncti
  9392. ons, including rejection.
  9393. However, very few probes could be confidently identified as differentially
  9394. methylated between healthy and dysfunctional transplants.
  9395. One likely explanation for this is the predominant influence of unobserved
  9396. confounding factors.
  9397. SVA can model and correct for such factors, but the correction can never
  9398. be perfect, so some degree of unwanted systematic variation will always
  9399. remain after SVA correction.
  9400. If the effect size of the confounding factors was similar to that of the
  9401. factor of interest (in this case, transplant status), this would be an
  9402. acceptable limitation, since removing most of the confounding factors'
  9403. effects would allow the main effect to stand out.
  9404. However, in this data set, the confounding factors have a much larger effect
  9405. size than transplant status, which means that the small degree of remaining
  9406. variation not removed by SVA can still swamp the effect of interest, making
  9407. it difficult to detect.
  9408. This is, of course, a major issue when the end goal is to develop a classifier
  9409. to diagnose transplant rejection from methylation data, since batch-correction
  9410. methods like SVA that work in a linear modeling context cannot be applied
  9411. in a machine learning context.
  9412. \end_layout
  9413. \begin_layout Standard
  9414. Currently, the source of these unwanted systematic variations in the data
  9415. is unknown.
  9416. The best solution would be to determine the cause of the variation and
  9417. eliminate it, thereby eliminating the need to model and remove that variation.
  9418. However, if this proves impractical, another option is to use SVA to identify
  9419. probes that are highly associated with the surrogate variables that describe
  9420. the unwanted variation in the data.
  9421. These probes could be discarded prior to classifier training, in order
  9422. to maximize the chance that the training algorithm will be able to identify
  9423. highly predictive probes from those remaining.
  9424. Lastly, it is possible that some of this unwanted variation is a result
  9425. of the array-based assay being used and would be eliminated by switching
  9426. to assaying DNA methylation using bisulphite sequencing.
  9427. However, this carries the risk that the sequencing assay will have its
  9428. own set of biases that must be corrected for in a different way.
  9429. \end_layout
  9430. \begin_layout Chapter
  9431. Globin-blocking for more effective blood RNA-seq analysis in primate animal
  9432. model
  9433. \end_layout
  9434. \begin_layout Standard
  9435. \begin_inset Flex TODO Note (inline)
  9436. status open
  9437. \begin_layout Plain Layout
  9438. Choose between above and the paper title: Optimizing yield of deep RNA sequencin
  9439. g for gene expression profiling by globin reduction of peripheral blood
  9440. samples from cynomolgus monkeys (Macaca fascicularis).
  9441. \end_layout
  9442. \end_inset
  9443. \end_layout
  9444. \begin_layout Standard
  9445. \begin_inset Flex TODO Note (inline)
  9446. status open
  9447. \begin_layout Plain Layout
  9448. Chapter author list: https://tex.stackexchange.com/questions/156862/displaying-aut
  9449. hor-for-each-chapter-in-book Every chapter gets an author list, which may
  9450. or may not be part of a citation to a published/preprinted paper.
  9451. \end_layout
  9452. \end_inset
  9453. \end_layout
  9454. \begin_layout Standard
  9455. \begin_inset Flex TODO Note (inline)
  9456. status open
  9457. \begin_layout Plain Layout
  9458. Preprint then cite the paper
  9459. \end_layout
  9460. \end_inset
  9461. \end_layout
  9462. \begin_layout Section*
  9463. Abstract
  9464. \end_layout
  9465. \begin_layout Paragraph
  9466. Background
  9467. \end_layout
  9468. \begin_layout Standard
  9469. Primate blood contains high concentrations of globin messenger RNA.
  9470. Globin reduction is a standard technique used to improve the expression
  9471. results obtained by DNA microarrays on RNA from blood samples.
  9472. However, with whole transcriptome RNA-sequencing (RNA-seq) quickly replacing
  9473. microarrays for many applications, the impact of globin reduction for RNA-seq
  9474. has not been previously studied.
  9475. Moreover, no off-the-shelf kits are available for globin reduction in nonhuman
  9476. primates.
  9477. \end_layout
  9478. \begin_layout Paragraph
  9479. Results
  9480. \end_layout
  9481. \begin_layout Standard
  9482. Here we report a protocol for RNA-seq in primate blood samples that uses
  9483. complimentary oligonucleotides to block reverse transcription of the alpha
  9484. and beta globin genes.
  9485. In test samples from cynomolgus monkeys (Macaca fascicularis), this globin
  9486. blocking protocol approximately doubles the yield of informative (non-globin)
  9487. reads by greatly reducing the fraction of globin reads, while also improving
  9488. the consistency in sequencing depth between samples.
  9489. The increased yield enables detection of about 2000 more genes, significantly
  9490. increases the correlation in measured gene expression levels between samples,
  9491. and increases the sensitivity of differential gene expression tests.
  9492. \end_layout
  9493. \begin_layout Paragraph
  9494. Conclusions
  9495. \end_layout
  9496. \begin_layout Standard
  9497. These results show that globin blocking significantly improves the cost-effectiv
  9498. eness of mRNA sequencing in primate blood samples by doubling the yield
  9499. of useful reads, allowing detection of more genes, and improving the precision
  9500. of gene expression measurements.
  9501. Based on these results, a globin reducing or blocking protocol is recommended
  9502. for all RNA-seq studies of primate blood samples.
  9503. \end_layout
  9504. \begin_layout Section
  9505. Approach
  9506. \end_layout
  9507. \begin_layout Standard
  9508. \begin_inset Note Note
  9509. status open
  9510. \begin_layout Plain Layout
  9511. Consider putting some of this in the Intro chapter
  9512. \end_layout
  9513. \begin_layout Itemize
  9514. Cynomolgus monkeys as a model organism
  9515. \end_layout
  9516. \begin_deeper
  9517. \begin_layout Itemize
  9518. Highly related to humans
  9519. \end_layout
  9520. \begin_layout Itemize
  9521. Small size and short life cycle - good research animal
  9522. \end_layout
  9523. \begin_layout Itemize
  9524. Genomics resources still in development
  9525. \end_layout
  9526. \end_deeper
  9527. \begin_layout Itemize
  9528. Inadequacy of existing blood RNA-seq protocols
  9529. \end_layout
  9530. \begin_deeper
  9531. \begin_layout Itemize
  9532. Existing protocols use a separate globin pulldown step, slowing down processing
  9533. \end_layout
  9534. \end_deeper
  9535. \end_inset
  9536. \end_layout
  9537. \begin_layout Standard
  9538. Increasingly, researchers are turning to high-throughput mRNA sequencing
  9539. technologies (RNA-seq) in preference to expression microarrays for analysis
  9540. of gene expression
  9541. \begin_inset CommandInset citation
  9542. LatexCommand cite
  9543. key "Mutz2012"
  9544. literal "false"
  9545. \end_inset
  9546. .
  9547. The advantages are even greater for study of model organisms with no well-estab
  9548. lished array platforms available, such as the cynomolgus monkey (Macaca
  9549. fascicularis).
  9550. High fractions of globin mRNA are naturally present in mammalian peripheral
  9551. blood samples (up to 70% of total mRNA) and these are known to interfere
  9552. with the results of array-based expression profiling
  9553. \begin_inset CommandInset citation
  9554. LatexCommand cite
  9555. key "Winn2010"
  9556. literal "false"
  9557. \end_inset
  9558. .
  9559. The importance of globin reduction for RNA-seq of blood has only been evaluated
  9560. for a deepSAGE protocol on human samples
  9561. \begin_inset CommandInset citation
  9562. LatexCommand cite
  9563. key "Mastrokolias2012"
  9564. literal "false"
  9565. \end_inset
  9566. .
  9567. In the present report, we evaluated globin reduction using custom blocking
  9568. oligonucleotides for deep RNA-seq of peripheral blood samples from a nonhuman
  9569. primate, cynomolgus monkey, using the Illumina technology platform.
  9570. We demonstrate that globin reduction significantly improves the cost-effectiven
  9571. ess of RNA-seq in blood samples.
  9572. Thus, our protocol offers a significant advantage to any investigator planning
  9573. to use RNA-seq for gene expression profiling of nonhuman primate blood
  9574. samples.
  9575. Our method can be generally applied to any species by designing complementary
  9576. oligonucleotide blocking probes to the globin gene sequences of that species.
  9577. Indeed, any highly expressed but biologically uninformative transcripts
  9578. can also be blocked to further increase sequencing efficiency and value
  9579. \begin_inset CommandInset citation
  9580. LatexCommand cite
  9581. key "Arnaud2016"
  9582. literal "false"
  9583. \end_inset
  9584. .
  9585. \end_layout
  9586. \begin_layout Section
  9587. Methods
  9588. \end_layout
  9589. \begin_layout Subsection
  9590. Sample collection
  9591. \end_layout
  9592. \begin_layout Standard
  9593. All research reported here was done under IACUC-approved protocols at the
  9594. University of Miami and complied with all applicable federal and state
  9595. regulations and ethical principles for nonhuman primate research.
  9596. Blood draws occurred between 16 April 2012 and 18 June 2015.
  9597. The experimental system involved intrahepatic pancreatic islet transplantation
  9598. into Cynomolgus monkeys with induced diabetes mellitus with or without
  9599. concomitant infusion of mesenchymal stem cells.
  9600. Blood was collected at serial time points before and after transplantation
  9601. into PAXgene Blood RNA tubes (PreAnalytiX/Qiagen, Valencia, CA) at the
  9602. precise volume:volume ratio of 2.5 ml whole blood into 6.9 ml of PAX gene
  9603. additive.
  9604. \end_layout
  9605. \begin_layout Subsection
  9606. Globin Blocking
  9607. \end_layout
  9608. \begin_layout Standard
  9609. Four oligonucleotides were designed to hybridize to the 3’ end of the transcript
  9610. s for Cynomolgus HBA1, HBA2 and HBB, with two hybridization sites for HBB
  9611. and 2 sites for HBA (the chosen sites were identical in both HBA genes).
  9612. All oligos were purchased from Sigma and were entirely composed of 2’O-Me
  9613. bases with a C3 spacer positioned at the 3’ ends to prevent any polymerase
  9614. mediated primer extension.
  9615. \end_layout
  9616. \begin_layout Quote
  9617. HBA1/2 site 1: GCCCACUCAGACUUUAUUCAAAG-C3spacer
  9618. \end_layout
  9619. \begin_layout Quote
  9620. HBA1/2 site 2: GGUGCAAGGAGGGGAGGAG-C3spacer
  9621. \end_layout
  9622. \begin_layout Quote
  9623. HBB site 1: AAUGAAAAUAAAUGUUUUUUAUUAG-C3spacer
  9624. \end_layout
  9625. \begin_layout Quote
  9626. HBB site 2: CUCAAGGCCCUUCAUAAUAUCCC-C3spacer
  9627. \end_layout
  9628. \begin_layout Subsection
  9629. RNA-seq Library Preparation
  9630. \end_layout
  9631. \begin_layout Standard
  9632. Sequencing libraries were prepared with 200ng total RNA from each sample.
  9633. Polyadenylated mRNA was selected from 200 ng aliquots of cynomologus blood-deri
  9634. ved total RNA using Ambion Dynabeads Oligo(dT)25 beads (Invitrogen) following
  9635. manufacturer’s recommended protocol.
  9636. PolyA selected RNA was then combined with 8 pmol of HBA1/2 (site 1), 8
  9637. pmol of HBA1/2 (site 2), 12 pmol of HBB (site 1) and 12 pmol of HBB (site
  9638. 2) oligonucleotides.
  9639. In addition, 20 pmol of RT primer containing a portion of the Illumina
  9640. adapter sequence (B-oligo-dTV: GAGTTCCTTGGCACCCGAGAATTCCATTTTTTTTTTTTTTTTTTTV)
  9641. and 4 µL of 5X First Strand buffer (250 mM Tris-HCl pH 8.3, 375 mM KCl,
  9642. 15mM MgCl2) were added in a total volume of 15 µL.
  9643. The RNA was fragmented by heating this cocktail for 3 minutes at 95°C and
  9644. then placed on ice.
  9645. This was followed by the addition of 2 µL 0.1 M DTT, 1 µL RNaseOUT, 1 µL
  9646. 10mM dNTPs 10% biotin-16 aminoallyl-2’- dUTP and 10% biotin-16 aminoallyl-2’-
  9647. dCTP (TriLink Biotech, San Diego, CA), 1 µL Superscript II (200U/ µL, Thermo-Fi
  9648. sher).
  9649. A second “unblocked” library was prepared in the same way for each sample
  9650. but replacing the blocking oligos with an equivalent volume of water.
  9651. The reaction was carried out at 25°C for 15 minutes and 42°C for 40 minutes,
  9652. followed by incubation at 75°C for 10 minutes to inactivate the reverse
  9653. transcriptase.
  9654. \end_layout
  9655. \begin_layout Standard
  9656. The cDNA/RNA hybrid molecules were purified using 1.8X Ampure XP beads (Agencourt
  9657. ) following supplier’s recommended protocol.
  9658. The cDNA/RNA hybrid was eluted in 25 µL of 10 mM Tris-HCl pH 8.0, and then
  9659. bound to 25 µL of M280 Magnetic Streptavidin beads washed per recommended
  9660. protocol (Thermo-Fisher).
  9661. After 30 minutes of binding, beads were washed one time in 100 µL 0.1N NaOH
  9662. to denature and remove the bound RNA, followed by two 100 µL washes with
  9663. 1X TE buffer.
  9664. \end_layout
  9665. \begin_layout Standard
  9666. Subsequent attachment of the 5-prime Illumina A adapter was performed by
  9667. on-bead random primer extension of the following sequence (A-N8 primer:
  9668. TTCAGAGTTCTACAGTCCGACGATCNNNNNNNN).
  9669. Briefly, beads were resuspended in a 20 µL reaction containing 5 µM A-N8
  9670. primer, 40mM Tris-HCl pH 7.5, 20mM MgCl2, 50mM NaCl, 0.325U/µL Sequenase
  9671. 2.0 (Affymetrix, Santa Clara, CA), 0.0025U/µL inorganic pyrophosphatase (Affymetr
  9672. ix) and 300 µM each dNTP.
  9673. Reaction was incubated at 22°C for 30 minutes, then beads were washed 2
  9674. times with 1X TE buffer (200µL).
  9675. \end_layout
  9676. \begin_layout Standard
  9677. The magnetic streptavidin beads were resuspended in 34 µL nuclease-free
  9678. water and added directly to a PCR tube.
  9679. The two Illumina protocol-specified PCR primers were added at 0.53 µM (Illumina
  9680. TruSeq Universal Primer 1 and Illumina TruSeq barcoded PCR primer 2), along
  9681. with 40 µL 2X KAPA HiFi Hotstart ReadyMix (KAPA, Willmington MA) and thermocycl
  9682. ed as follows: starting with 98°C (2 min-hold); 15 cycles of 98°C, 20sec;
  9683. 60°C, 30sec; 72°C, 30sec; and finished with a 72°C (2 min-hold).
  9684. \end_layout
  9685. \begin_layout Standard
  9686. PCR products were purified with 1X Ampure Beads following manufacturer’s
  9687. recommended protocol.
  9688. Libraries were then analyzed using the Agilent TapeStation and quantitation
  9689. of desired size range was performed by “smear analysis”.
  9690. Samples were pooled in equimolar batches of 16 samples.
  9691. Pooled libraries were size selected on 2% agarose gels (E-Gel EX Agarose
  9692. Gels; Thermo-Fisher).
  9693. Products were cut between 250 and 350 bp (corresponding to insert sizes
  9694. of 130 to 230 bps).
  9695. Finished library pools were then sequenced on the Illumina NextSeq500 instrumen
  9696. t with 75 base read lengths.
  9697. \end_layout
  9698. \begin_layout Subsection
  9699. Read alignment and counting
  9700. \end_layout
  9701. \begin_layout Standard
  9702. Reads were aligned to the cynomolgus genome using STAR
  9703. \begin_inset CommandInset citation
  9704. LatexCommand cite
  9705. key "Dobin2013,Wilson2013"
  9706. literal "false"
  9707. \end_inset
  9708. .
  9709. Counts of uniquely mapped reads were obtained for every gene in each sample
  9710. with the “featureCounts” function from the Rsubread package, using each
  9711. of the three possibilities for the “strandSpecific” option: sense, antisense,
  9712. and unstranded
  9713. \begin_inset CommandInset citation
  9714. LatexCommand cite
  9715. key "Liao2014"
  9716. literal "false"
  9717. \end_inset
  9718. .
  9719. A few artifacts in the cynomolgus genome annotation complicated read counting.
  9720. First, no ortholog is annotated for alpha globin in the cynomolgus genome,
  9721. presumably because the human genome has two alpha globin genes with nearly
  9722. identical sequences, making the orthology relationship ambiguous.
  9723. However, two loci in the cynomolgus genome are as “hemoglobin subunit alpha-lik
  9724. e” (LOC102136192 and LOC102136846).
  9725. LOC102136192 is annotated as a pseudogene while LOC102136846 is annotated
  9726. as protein-coding.
  9727. Our globin reduction protocol was designed to include blocking of these
  9728. two genes.
  9729. Indeed, these two genes have almost the same read counts in each library
  9730. as the properly-annotated HBB gene and much larger counts than any other
  9731. gene in the unblocked libraries, giving confidence that reads derived from
  9732. the real alpha globin are mapping to both genes.
  9733. Thus, reads from both of these loci were counted as alpha globin reads
  9734. in all further analyses.
  9735. The second artifact is a small, uncharacterized non-coding RNA gene (LOC1021365
  9736. 91), which overlaps the HBA-like gene (LOC102136192) on the opposite strand.
  9737. If counting is not performed in stranded mode (or if a non-strand-specific
  9738. sequencing protocol is used), many reads mapping to the globin gene will
  9739. be discarded as ambiguous due to their overlap with this ncRNA gene, resulting
  9740. in significant undercounting of globin reads.
  9741. Therefore, stranded sense counts were used for all further analysis in
  9742. the present study to insure that we accurately accounted for globin transcript
  9743. reduction.
  9744. However, we note that stranded reads are not necessary for RNA-seq using
  9745. our protocol in standard practice.
  9746. \end_layout
  9747. \begin_layout Subsection
  9748. Normalization and Exploratory Data Analysis
  9749. \end_layout
  9750. \begin_layout Standard
  9751. Libraries were normalized by computing scaling factors using the edgeR package’s
  9752. Trimmed Mean of M-values method
  9753. \begin_inset CommandInset citation
  9754. LatexCommand cite
  9755. key "Robinson2010"
  9756. literal "false"
  9757. \end_inset
  9758. .
  9759. Log2 counts per million values (logCPM) were calculated using the cpm function
  9760. in edgeR for individual samples and aveLogCPM function for averages across
  9761. groups of samples, using those functions’ default prior count values to
  9762. avoid taking the logarithm of 0.
  9763. Genes were considered “present” if their average normalized logCPM values
  9764. across all libraries were at least -1.
  9765. Normalizing for gene length was unnecessary because the sequencing protocol
  9766. is 3’-biased and hence the expected read count for each gene is related
  9767. to the transcript’s copy number but not its length.
  9768. \end_layout
  9769. \begin_layout Standard
  9770. In order to assess the effect of blocking on reproducibility, Pearson and
  9771. Spearman correlation coefficients were computed between the logCPM values
  9772. for every pair of libraries within the globin-blocked (GB) and unblocked
  9773. (non-GB) groups, and edgeR's “estimateDisp” function was used to compute
  9774. negative binomial dispersions separately for the two groups
  9775. \begin_inset CommandInset citation
  9776. LatexCommand cite
  9777. key "Chen2014"
  9778. literal "false"
  9779. \end_inset
  9780. .
  9781. \end_layout
  9782. \begin_layout Subsection
  9783. Differential Expression Analysis
  9784. \end_layout
  9785. \begin_layout Standard
  9786. All tests for differential gene expression were performed using edgeR, by
  9787. first fitting a negative binomial generalized linear model to the counts
  9788. and normalization factors and then performing a quasi-likelihood F-test
  9789. with robust estimation of outlier gene dispersions
  9790. \begin_inset CommandInset citation
  9791. LatexCommand cite
  9792. key "Lund2012,Phipson2016"
  9793. literal "false"
  9794. \end_inset
  9795. .
  9796. To investigate the effects of globin blocking on each gene, an additive
  9797. model was fit to the full data with coefficients for globin blocking and
  9798. SampleID.
  9799. To test the effect of globin blocking on detection of differentially expressed
  9800. genes, the GB samples and non-GB samples were each analyzed independently
  9801. as follows: for each animal with both a pre-transplant and a post-transplant
  9802. time point in the data set, the pre-transplant sample and the earliest
  9803. post-transplant sample were selected, and all others were excluded, yielding
  9804. a pre-/post-transplant pair of samples for each animal (N=7 animals with
  9805. paired samples).
  9806. These samples were analyzed for pre-transplant vs.
  9807. post-transplant differential gene expression while controlling for inter-animal
  9808. variation using an additive model with coefficients for transplant and
  9809. animal ID.
  9810. In all analyses, p-values were adjusted using the Benjamini-Hochberg procedure
  9811. for FDR control
  9812. \begin_inset CommandInset citation
  9813. LatexCommand cite
  9814. key "Benjamini1995"
  9815. literal "false"
  9816. \end_inset
  9817. .
  9818. \end_layout
  9819. \begin_layout Standard
  9820. \begin_inset Note Note
  9821. status open
  9822. \begin_layout Itemize
  9823. New blood RNA-seq protocol to block reverse transcription of globin genes
  9824. \end_layout
  9825. \begin_layout Itemize
  9826. Blood RNA-seq time course after transplants with/without MSC infusion
  9827. \end_layout
  9828. \end_inset
  9829. \end_layout
  9830. \begin_layout Section
  9831. Results
  9832. \end_layout
  9833. \begin_layout Subsection
  9834. Globin blocking yields a larger and more consistent fraction of useful reads
  9835. \end_layout
  9836. \begin_layout Standard
  9837. \begin_inset ERT
  9838. status open
  9839. \begin_layout Plain Layout
  9840. \backslash
  9841. afterpage{
  9842. \end_layout
  9843. \begin_layout Plain Layout
  9844. \backslash
  9845. begin{landscape}
  9846. \end_layout
  9847. \end_inset
  9848. \end_layout
  9849. \begin_layout Standard
  9850. \begin_inset Float table
  9851. placement p
  9852. wide false
  9853. sideways false
  9854. status open
  9855. \begin_layout Plain Layout
  9856. \align center
  9857. \begin_inset Tabular
  9858. <lyxtabular version="3" rows="4" columns="7">
  9859. <features tabularvalignment="middle">
  9860. <column alignment="center" valignment="top">
  9861. <column alignment="center" valignment="top">
  9862. <column alignment="center" valignment="top">
  9863. <column alignment="center" valignment="top">
  9864. <column alignment="center" valignment="top">
  9865. <column alignment="center" valignment="top">
  9866. <column alignment="center" valignment="top">
  9867. <row>
  9868. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  9869. \begin_inset Text
  9870. \begin_layout Plain Layout
  9871. \end_layout
  9872. \end_inset
  9873. </cell>
  9874. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  9875. \begin_inset Text
  9876. \begin_layout Plain Layout
  9877. \family roman
  9878. \series medium
  9879. \shape up
  9880. \size normal
  9881. \emph off
  9882. \bar no
  9883. \strikeout off
  9884. \xout off
  9885. \uuline off
  9886. \uwave off
  9887. \noun off
  9888. \color none
  9889. Percent of Total Reads
  9890. \end_layout
  9891. \end_inset
  9892. </cell>
  9893. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  9894. \begin_inset Text
  9895. \begin_layout Plain Layout
  9896. \end_layout
  9897. \end_inset
  9898. </cell>
  9899. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  9900. \begin_inset Text
  9901. \begin_layout Plain Layout
  9902. \end_layout
  9903. \end_inset
  9904. </cell>
  9905. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  9906. \begin_inset Text
  9907. \begin_layout Plain Layout
  9908. \end_layout
  9909. \end_inset
  9910. </cell>
  9911. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  9912. \begin_inset Text
  9913. \begin_layout Plain Layout
  9914. \family roman
  9915. \series medium
  9916. \shape up
  9917. \size normal
  9918. \emph off
  9919. \bar no
  9920. \strikeout off
  9921. \xout off
  9922. \uuline off
  9923. \uwave off
  9924. \noun off
  9925. \color none
  9926. Percent of Genic Reads
  9927. \end_layout
  9928. \end_inset
  9929. </cell>
  9930. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  9931. \begin_inset Text
  9932. \begin_layout Plain Layout
  9933. \end_layout
  9934. \end_inset
  9935. </cell>
  9936. </row>
  9937. <row>
  9938. <cell alignment="center" valignment="top" bottomline="true" leftline="true" usebox="none">
  9939. \begin_inset Text
  9940. \begin_layout Plain Layout
  9941. GB
  9942. \end_layout
  9943. \end_inset
  9944. </cell>
  9945. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  9946. \begin_inset Text
  9947. \begin_layout Plain Layout
  9948. \family roman
  9949. \series medium
  9950. \shape up
  9951. \size normal
  9952. \emph off
  9953. \bar no
  9954. \strikeout off
  9955. \xout off
  9956. \uuline off
  9957. \uwave off
  9958. \noun off
  9959. \color none
  9960. Non-globin Reads
  9961. \end_layout
  9962. \end_inset
  9963. </cell>
  9964. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  9965. \begin_inset Text
  9966. \begin_layout Plain Layout
  9967. \family roman
  9968. \series medium
  9969. \shape up
  9970. \size normal
  9971. \emph off
  9972. \bar no
  9973. \strikeout off
  9974. \xout off
  9975. \uuline off
  9976. \uwave off
  9977. \noun off
  9978. \color none
  9979. Globin Reads
  9980. \end_layout
  9981. \end_inset
  9982. </cell>
  9983. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  9984. \begin_inset Text
  9985. \begin_layout Plain Layout
  9986. \family roman
  9987. \series medium
  9988. \shape up
  9989. \size normal
  9990. \emph off
  9991. \bar no
  9992. \strikeout off
  9993. \xout off
  9994. \uuline off
  9995. \uwave off
  9996. \noun off
  9997. \color none
  9998. All Genic Reads
  9999. \end_layout
  10000. \end_inset
  10001. </cell>
  10002. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10003. \begin_inset Text
  10004. \begin_layout Plain Layout
  10005. \family roman
  10006. \series medium
  10007. \shape up
  10008. \size normal
  10009. \emph off
  10010. \bar no
  10011. \strikeout off
  10012. \xout off
  10013. \uuline off
  10014. \uwave off
  10015. \noun off
  10016. \color none
  10017. All Aligned Reads
  10018. \end_layout
  10019. \end_inset
  10020. </cell>
  10021. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10022. \begin_inset Text
  10023. \begin_layout Plain Layout
  10024. \family roman
  10025. \series medium
  10026. \shape up
  10027. \size normal
  10028. \emph off
  10029. \bar no
  10030. \strikeout off
  10031. \xout off
  10032. \uuline off
  10033. \uwave off
  10034. \noun off
  10035. \color none
  10036. Non-globin Reads
  10037. \end_layout
  10038. \end_inset
  10039. </cell>
  10040. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  10041. \begin_inset Text
  10042. \begin_layout Plain Layout
  10043. \family roman
  10044. \series medium
  10045. \shape up
  10046. \size normal
  10047. \emph off
  10048. \bar no
  10049. \strikeout off
  10050. \xout off
  10051. \uuline off
  10052. \uwave off
  10053. \noun off
  10054. \color none
  10055. Globin Reads
  10056. \end_layout
  10057. \end_inset
  10058. </cell>
  10059. </row>
  10060. <row>
  10061. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10062. \begin_inset Text
  10063. \begin_layout Plain Layout
  10064. \family roman
  10065. \series medium
  10066. \shape up
  10067. \size normal
  10068. \emph off
  10069. \bar no
  10070. \strikeout off
  10071. \xout off
  10072. \uuline off
  10073. \uwave off
  10074. \noun off
  10075. \color none
  10076. Yes
  10077. \end_layout
  10078. \end_inset
  10079. </cell>
  10080. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10081. \begin_inset Text
  10082. \begin_layout Plain Layout
  10083. \family roman
  10084. \series medium
  10085. \shape up
  10086. \size normal
  10087. \emph off
  10088. \bar no
  10089. \strikeout off
  10090. \xout off
  10091. \uuline off
  10092. \uwave off
  10093. \noun off
  10094. \color none
  10095. 50.4% ± 6.82
  10096. \end_layout
  10097. \end_inset
  10098. </cell>
  10099. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10100. \begin_inset Text
  10101. \begin_layout Plain Layout
  10102. \family roman
  10103. \series medium
  10104. \shape up
  10105. \size normal
  10106. \emph off
  10107. \bar no
  10108. \strikeout off
  10109. \xout off
  10110. \uuline off
  10111. \uwave off
  10112. \noun off
  10113. \color none
  10114. 3.48% ± 2.94
  10115. \end_layout
  10116. \end_inset
  10117. </cell>
  10118. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10119. \begin_inset Text
  10120. \begin_layout Plain Layout
  10121. \family roman
  10122. \series medium
  10123. \shape up
  10124. \size normal
  10125. \emph off
  10126. \bar no
  10127. \strikeout off
  10128. \xout off
  10129. \uuline off
  10130. \uwave off
  10131. \noun off
  10132. \color none
  10133. 53.9% ± 6.81
  10134. \end_layout
  10135. \end_inset
  10136. </cell>
  10137. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10138. \begin_inset Text
  10139. \begin_layout Plain Layout
  10140. \family roman
  10141. \series medium
  10142. \shape up
  10143. \size normal
  10144. \emph off
  10145. \bar no
  10146. \strikeout off
  10147. \xout off
  10148. \uuline off
  10149. \uwave off
  10150. \noun off
  10151. \color none
  10152. 89.7% ± 2.40
  10153. \end_layout
  10154. \end_inset
  10155. </cell>
  10156. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10157. \begin_inset Text
  10158. \begin_layout Plain Layout
  10159. \family roman
  10160. \series medium
  10161. \shape up
  10162. \size normal
  10163. \emph off
  10164. \bar no
  10165. \strikeout off
  10166. \xout off
  10167. \uuline off
  10168. \uwave off
  10169. \noun off
  10170. \color none
  10171. 93.5% ± 5.25
  10172. \end_layout
  10173. \end_inset
  10174. </cell>
  10175. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  10176. \begin_inset Text
  10177. \begin_layout Plain Layout
  10178. \family roman
  10179. \series medium
  10180. \shape up
  10181. \size normal
  10182. \emph off
  10183. \bar no
  10184. \strikeout off
  10185. \xout off
  10186. \uuline off
  10187. \uwave off
  10188. \noun off
  10189. \color none
  10190. 6.49% ± 5.25
  10191. \end_layout
  10192. \end_inset
  10193. </cell>
  10194. </row>
  10195. <row>
  10196. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10197. \begin_inset Text
  10198. \begin_layout Plain Layout
  10199. \family roman
  10200. \series medium
  10201. \shape up
  10202. \size normal
  10203. \emph off
  10204. \bar no
  10205. \strikeout off
  10206. \xout off
  10207. \uuline off
  10208. \uwave off
  10209. \noun off
  10210. \color none
  10211. No
  10212. \end_layout
  10213. \end_inset
  10214. </cell>
  10215. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10216. \begin_inset Text
  10217. \begin_layout Plain Layout
  10218. \family roman
  10219. \series medium
  10220. \shape up
  10221. \size normal
  10222. \emph off
  10223. \bar no
  10224. \strikeout off
  10225. \xout off
  10226. \uuline off
  10227. \uwave off
  10228. \noun off
  10229. \color none
  10230. 26.3% ± 8.95
  10231. \end_layout
  10232. \end_inset
  10233. </cell>
  10234. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10235. \begin_inset Text
  10236. \begin_layout Plain Layout
  10237. \family roman
  10238. \series medium
  10239. \shape up
  10240. \size normal
  10241. \emph off
  10242. \bar no
  10243. \strikeout off
  10244. \xout off
  10245. \uuline off
  10246. \uwave off
  10247. \noun off
  10248. \color none
  10249. 44.6% ± 16.6
  10250. \end_layout
  10251. \end_inset
  10252. </cell>
  10253. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10254. \begin_inset Text
  10255. \begin_layout Plain Layout
  10256. \family roman
  10257. \series medium
  10258. \shape up
  10259. \size normal
  10260. \emph off
  10261. \bar no
  10262. \strikeout off
  10263. \xout off
  10264. \uuline off
  10265. \uwave off
  10266. \noun off
  10267. \color none
  10268. 70.1% ± 9.38
  10269. \end_layout
  10270. \end_inset
  10271. </cell>
  10272. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10273. \begin_inset Text
  10274. \begin_layout Plain Layout
  10275. \family roman
  10276. \series medium
  10277. \shape up
  10278. \size normal
  10279. \emph off
  10280. \bar no
  10281. \strikeout off
  10282. \xout off
  10283. \uuline off
  10284. \uwave off
  10285. \noun off
  10286. \color none
  10287. 90.7% ± 5.16
  10288. \end_layout
  10289. \end_inset
  10290. </cell>
  10291. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10292. \begin_inset Text
  10293. \begin_layout Plain Layout
  10294. \family roman
  10295. \series medium
  10296. \shape up
  10297. \size normal
  10298. \emph off
  10299. \bar no
  10300. \strikeout off
  10301. \xout off
  10302. \uuline off
  10303. \uwave off
  10304. \noun off
  10305. \color none
  10306. 38.8% ± 17.1
  10307. \end_layout
  10308. \end_inset
  10309. </cell>
  10310. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  10311. \begin_inset Text
  10312. \begin_layout Plain Layout
  10313. \family roman
  10314. \series medium
  10315. \shape up
  10316. \size normal
  10317. \emph off
  10318. \bar no
  10319. \strikeout off
  10320. \xout off
  10321. \uuline off
  10322. \uwave off
  10323. \noun off
  10324. \color none
  10325. 61.2% ± 17.1
  10326. \end_layout
  10327. \end_inset
  10328. </cell>
  10329. </row>
  10330. </lyxtabular>
  10331. \end_inset
  10332. \end_layout
  10333. \begin_layout Plain Layout
  10334. \begin_inset Caption Standard
  10335. \begin_layout Plain Layout
  10336. \series bold
  10337. \begin_inset Argument 1
  10338. status collapsed
  10339. \begin_layout Plain Layout
  10340. Fractions of reads mapping to genomic features in GB and non-GB samples.
  10341. \end_layout
  10342. \end_inset
  10343. \begin_inset CommandInset label
  10344. LatexCommand label
  10345. name "tab:Fractions-of-reads"
  10346. \end_inset
  10347. Fractions of reads mapping to genomic features in GB and non-GB samples.
  10348. \series default
  10349. All values are given as mean ± standard deviation.
  10350. \end_layout
  10351. \end_inset
  10352. \end_layout
  10353. \end_inset
  10354. \end_layout
  10355. \begin_layout Standard
  10356. \begin_inset ERT
  10357. status open
  10358. \begin_layout Plain Layout
  10359. \backslash
  10360. end{landscape}
  10361. \end_layout
  10362. \begin_layout Plain Layout
  10363. }
  10364. \end_layout
  10365. \end_inset
  10366. \end_layout
  10367. \begin_layout Standard
  10368. The objective of the present study was to validate a new protocol for deep
  10369. RNA-seq of whole blood drawn into PaxGene tubes from cynomolgus monkeys
  10370. undergoing islet transplantation, with particular focus on minimizing the
  10371. loss of useful sequencing space to uninformative globin reads.
  10372. The details of the analysis with respect to transplant outcomes and the
  10373. impact of mesenchymal stem cell treatment will be reported in a separate
  10374. manuscript (in preparation).
  10375. To focus on the efficacy of our globin blocking protocol, 37 blood samples,
  10376. 16 from pre-transplant and 21 from post-transplant time points, were each
  10377. prepped once with and once without globin blocking oligos, and were then
  10378. sequenced on an Illumina NextSeq500 instrument.
  10379. The number of reads aligning to each gene in the cynomolgus genome was
  10380. counted.
  10381. Table 1 summarizes the distribution of read fractions among the GB and
  10382. non-GB libraries.
  10383. In the libraries with no globin blocking, globin reads made up an average
  10384. of 44.6% of total input reads, while reads assigned to all other genes made
  10385. up an average of 26.3%.
  10386. The remaining reads either aligned to intergenic regions (that include
  10387. long non-coding RNAs) or did not align with any annotated transcripts in
  10388. the current build of the cynomolgus genome.
  10389. In the GB libraries, globin reads made up only 3.48% and reads assigned
  10390. to all other genes increased to 50.4%.
  10391. Thus, globin blocking resulted in a 92.2% reduction in globin reads and
  10392. a 91.6% increase in yield of useful non-globin reads.
  10393. \end_layout
  10394. \begin_layout Standard
  10395. This reduction is not quite as efficient as the previous analysis showed
  10396. for human samples by DeepSAGE (<0.4% globin reads after globin reduction)
  10397. \begin_inset CommandInset citation
  10398. LatexCommand cite
  10399. key "Mastrokolias2012"
  10400. literal "false"
  10401. \end_inset
  10402. .
  10403. Nonetheless, this degree of globin reduction is sufficient to nearly double
  10404. the yield of useful reads.
  10405. Thus, globin blocking cuts the required sequencing effort (and costs) to
  10406. achieve a target coverage depth by almost 50%.
  10407. Consistent with this near doubling of yield, the average difference in
  10408. un-normalized logCPM across all genes between the GB libraries and non-GB
  10409. libraries is approximately 1 (mean = 1.01, median = 1.08), an overall 2-fold
  10410. increase.
  10411. Un-normalized values are used here because the TMM normalization correctly
  10412. identifies this 2-fold difference as biologically irrelevant and removes
  10413. it.
  10414. \end_layout
  10415. \begin_layout Standard
  10416. \begin_inset Float figure
  10417. wide false
  10418. sideways false
  10419. status collapsed
  10420. \begin_layout Plain Layout
  10421. \align center
  10422. \begin_inset Graphics
  10423. filename graphics/Globin Paper/figure1 - globin-fractions.pdf
  10424. lyxscale 50
  10425. width 75col%
  10426. \end_inset
  10427. \end_layout
  10428. \begin_layout Plain Layout
  10429. \begin_inset Caption Standard
  10430. \begin_layout Plain Layout
  10431. \series bold
  10432. \begin_inset Argument 1
  10433. status collapsed
  10434. \begin_layout Plain Layout
  10435. Fraction of genic reads in each sample aligned to non-globin genes, with
  10436. and without globin blocking (GB).
  10437. \end_layout
  10438. \end_inset
  10439. \begin_inset CommandInset label
  10440. LatexCommand label
  10441. name "fig:Fraction-of-genic-reads"
  10442. \end_inset
  10443. Fraction of genic reads in each sample aligned to non-globin genes, with
  10444. and without globin blocking (GB).
  10445. \series default
  10446. All reads in each sequencing library were aligned to the cyno genome, and
  10447. the number of reads uniquely aligning to each gene was counted.
  10448. For each sample, counts were summed separately for all globin genes and
  10449. for the remainder of the genes (non-globin genes), and the fraction of
  10450. genic reads aligned to non-globin genes was computed.
  10451. Each point represents an individual sample.
  10452. Gray + signs indicate the means for globin-blocked libraries and unblocked
  10453. libraries.
  10454. The overall distribution for each group is represented as a notched box
  10455. plots.
  10456. Points are randomly spread vertically to avoid excessive overlapping.
  10457. \end_layout
  10458. \end_inset
  10459. \end_layout
  10460. \end_inset
  10461. \end_layout
  10462. \begin_layout Standard
  10463. Another important aspect is that the standard deviations in Table
  10464. \begin_inset CommandInset ref
  10465. LatexCommand ref
  10466. reference "tab:Fractions-of-reads"
  10467. plural "false"
  10468. caps "false"
  10469. noprefix "false"
  10470. \end_inset
  10471. are uniformly smaller in the GB samples than the non-GB ones, indicating
  10472. much greater consistency of yield.
  10473. This is best seen in the percentage of non-globin reads as a fraction of
  10474. total reads aligned to annotated genes (genic reads).
  10475. For the non-GB samples, this measure ranges from 10.9% to 80.9%, while for
  10476. the GB samples it ranges from 81.9% to 99.9% (Figure
  10477. \begin_inset CommandInset ref
  10478. LatexCommand ref
  10479. reference "fig:Fraction-of-genic-reads"
  10480. plural "false"
  10481. caps "false"
  10482. noprefix "false"
  10483. \end_inset
  10484. ).
  10485. This means that for applications where it is critical that each sample
  10486. achieve a specified minimum coverage in order to provide useful information,
  10487. it would be necessary to budget up to 10 times the sequencing depth per
  10488. sample without globin blocking, even though the average yield improvement
  10489. for globin blocking is only 2-fold, because every sample has a chance of
  10490. being 90% globin and 10% useful reads.
  10491. Hence, the more consistent behavior of GB samples makes planning an experiment
  10492. easier and more efficient because it eliminates the need to over-sequence
  10493. every sample in order to guard against the worst case of a high-globin
  10494. fraction.
  10495. \end_layout
  10496. \begin_layout Subsection
  10497. Globin blocking lowers the noise floor and allows detection of about 2000
  10498. more low-expression genes
  10499. \end_layout
  10500. \begin_layout Standard
  10501. \begin_inset Flex TODO Note (inline)
  10502. status open
  10503. \begin_layout Plain Layout
  10504. Remove redundant titles from figures
  10505. \end_layout
  10506. \end_inset
  10507. \end_layout
  10508. \begin_layout Standard
  10509. \begin_inset Float figure
  10510. wide false
  10511. sideways false
  10512. status collapsed
  10513. \begin_layout Plain Layout
  10514. \align center
  10515. \begin_inset Graphics
  10516. filename graphics/Globin Paper/figure2 - aveLogCPM-colored.pdf
  10517. lyxscale 50
  10518. height 60theight%
  10519. \end_inset
  10520. \end_layout
  10521. \begin_layout Plain Layout
  10522. \begin_inset Caption Standard
  10523. \begin_layout Plain Layout
  10524. \series bold
  10525. \begin_inset Argument 1
  10526. status collapsed
  10527. \begin_layout Plain Layout
  10528. Distributions of average group gene abundances when normalized separately
  10529. or together.
  10530. \end_layout
  10531. \end_inset
  10532. \begin_inset CommandInset label
  10533. LatexCommand label
  10534. name "fig:logcpm-dists"
  10535. \end_inset
  10536. Distributions of average group gene abundances when normalized separately
  10537. or together.
  10538. \series default
  10539. All reads in each sequencing library were aligned to the cyno genome, and
  10540. the number of reads uniquely aligning to each gene was counted.
  10541. Genes with zero counts in all libraries were discarded.
  10542. Libraries were normalized using the TMM method.
  10543. Libraries were split into globin-blocked (GB) and non-GB groups and the
  10544. average abundance for each gene in both groups, measured in log2 counts
  10545. per million reads counted, was computed using the aveLogCPM function.
  10546. The distribution of average gene logCPM values was plotted for both groups
  10547. using a kernel density plot to approximate a continuous distribution.
  10548. The logCPM GB distributions are marked in red, non-GB in blue.
  10549. The black vertical line denotes the chosen detection threshold of -1.
  10550. Top panel: Libraries were split into GB and non-GB groups first and normalized
  10551. separately.
  10552. Bottom panel: Libraries were all normalized together first and then split
  10553. into groups.
  10554. \end_layout
  10555. \end_inset
  10556. \end_layout
  10557. \begin_layout Plain Layout
  10558. \end_layout
  10559. \end_inset
  10560. \end_layout
  10561. \begin_layout Standard
  10562. Since globin blocking yields more usable sequencing depth, it should also
  10563. allow detection of more genes at any given threshold.
  10564. When we looked at the distribution of average normalized logCPM values
  10565. across all libraries for genes with at least one read assigned to them,
  10566. we observed the expected bimodal distribution, with a high-abundance "signal"
  10567. peak representing detected genes and a low-abundance "noise" peak representing
  10568. genes whose read count did not rise above the noise floor (Figure
  10569. \begin_inset CommandInset ref
  10570. LatexCommand ref
  10571. reference "fig:logcpm-dists"
  10572. plural "false"
  10573. caps "false"
  10574. noprefix "false"
  10575. \end_inset
  10576. ).
  10577. Consistent with the 2-fold increase in raw counts assigned to non-globin
  10578. genes, the signal peak for GB samples is shifted to the right relative
  10579. to the non-GB signal peak.
  10580. When all the samples are normalized together, this difference is normalized
  10581. out, lining up the signal peaks, and this reveals that, as expected, the
  10582. noise floor for the GB samples is about 2-fold lower.
  10583. This greater separation between signal and noise peaks in the GB samples
  10584. means that low-expression genes should be more easily detected and more
  10585. precisely quantified than in the non-GB samples.
  10586. \end_layout
  10587. \begin_layout Standard
  10588. \begin_inset Float figure
  10589. wide false
  10590. sideways false
  10591. status collapsed
  10592. \begin_layout Plain Layout
  10593. \align center
  10594. \begin_inset Graphics
  10595. filename graphics/Globin Paper/figure3 - detection.pdf
  10596. lyxscale 50
  10597. width 70col%
  10598. \end_inset
  10599. \end_layout
  10600. \begin_layout Plain Layout
  10601. \begin_inset Caption Standard
  10602. \begin_layout Plain Layout
  10603. \series bold
  10604. \begin_inset Argument 1
  10605. status collapsed
  10606. \begin_layout Plain Layout
  10607. Gene detections as a function of abundance thresholds in globin-blocked
  10608. (GB) and non-GB samples.
  10609. \end_layout
  10610. \end_inset
  10611. \begin_inset CommandInset label
  10612. LatexCommand label
  10613. name "fig:Gene-detections"
  10614. \end_inset
  10615. Gene detections as a function of abundance thresholds in globin-blocked
  10616. (GB) and non-GB samples.
  10617. \series default
  10618. Average abundance (logCPM,
  10619. \begin_inset Formula $\log_{2}$
  10620. \end_inset
  10621. counts per million reads counted) was computed by separate group normalization
  10622. as described in Figure
  10623. \begin_inset CommandInset ref
  10624. LatexCommand ref
  10625. reference "fig:logcpm-dists"
  10626. plural "false"
  10627. caps "false"
  10628. noprefix "false"
  10629. \end_inset
  10630. for both the GB and non-GB groups, as well as for all samples considered
  10631. as one large group.
  10632. For each every integer threshold from -2 to 3, the number of genes detected
  10633. at or above that logCPM threshold was plotted for each group.
  10634. \end_layout
  10635. \end_inset
  10636. \end_layout
  10637. \begin_layout Plain Layout
  10638. \end_layout
  10639. \end_inset
  10640. \end_layout
  10641. \begin_layout Standard
  10642. Based on these distributions, we selected a detection threshold of -1, which
  10643. is approximately the leftmost edge of the trough between the signal and
  10644. noise peaks.
  10645. This represents the most liberal possible detection threshold that doesn't
  10646. call substantial numbers of noise genes as detected.
  10647. Among the full dataset, 13429 genes were detected at this threshold, and
  10648. 22276 were not.
  10649. When considering the GB libraries and non-GB libraries separately and re-comput
  10650. ing normalization factors independently within each group, 14535 genes were
  10651. detected in the GB libraries while only 12460 were detected in the non-GB
  10652. libraries.
  10653. Thus, GB allowed the detection of 2000 extra genes that were buried under
  10654. the noise floor without GB.
  10655. This pattern of at least 2000 additional genes detected with GB was also
  10656. consistent across a wide range of possible detection thresholds, from -2
  10657. to 3 (see Figure
  10658. \begin_inset CommandInset ref
  10659. LatexCommand ref
  10660. reference "fig:Gene-detections"
  10661. plural "false"
  10662. caps "false"
  10663. noprefix "false"
  10664. \end_inset
  10665. ).
  10666. \end_layout
  10667. \begin_layout Subsection
  10668. Globin blocking does not add significant additional noise or decrease sample
  10669. quality
  10670. \end_layout
  10671. \begin_layout Standard
  10672. One potential worry is that the globin blocking protocol could perturb the
  10673. levels of non-globin genes.
  10674. There are two kinds of possible perturbations: systematic and random.
  10675. The former is not a major concern for detection of differential expression,
  10676. since a 2-fold change in every sample has no effect on the relative fold
  10677. change between samples.
  10678. In contrast, random perturbations would increase the noise and obscure
  10679. the signal in the dataset, reducing the capacity to detect differential
  10680. expression.
  10681. \end_layout
  10682. \begin_layout Standard
  10683. \begin_inset Float figure
  10684. wide false
  10685. sideways false
  10686. status collapsed
  10687. \begin_layout Plain Layout
  10688. \align center
  10689. \begin_inset Graphics
  10690. filename graphics/Globin Paper/figure4 - maplot-colored.pdf
  10691. lyxscale 50
  10692. width 60col%
  10693. groupId colwidth
  10694. \end_inset
  10695. \end_layout
  10696. \begin_layout Plain Layout
  10697. \begin_inset Caption Standard
  10698. \begin_layout Plain Layout
  10699. \begin_inset Argument 1
  10700. status collapsed
  10701. \begin_layout Plain Layout
  10702. MA plot showing effects of globin blocking on each gene's abundance.
  10703. \end_layout
  10704. \end_inset
  10705. \begin_inset CommandInset label
  10706. LatexCommand label
  10707. name "fig:MA-plot"
  10708. \end_inset
  10709. \series bold
  10710. MA plot showing effects of globin blocking on each gene's abundance.
  10711. \series default
  10712. All libraries were normalized together as described in Figure
  10713. \begin_inset CommandInset ref
  10714. LatexCommand ref
  10715. reference "fig:logcpm-dists"
  10716. plural "false"
  10717. caps "false"
  10718. noprefix "false"
  10719. \end_inset
  10720. , and genes with an average logCPM below -1 were filtered out.
  10721. Each remaining gene was tested for differential abundance with respect
  10722. to globin blocking (GB) using edgeR’s quasi-likelihood F-test, fitting
  10723. a negative binomial generalized linear model to table of read counts in
  10724. each library.
  10725. For each gene, edgeR reported average abundance (logCPM),
  10726. \begin_inset Formula $\log_{2}$
  10727. \end_inset
  10728. fold change (logFC), p-value, and Benjamini-Hochberg adjusted false discovery
  10729. rate (FDR).
  10730. Each gene's logFC was plotted against its logCPM, colored by FDR.
  10731. Red points are significant at ≤10% FDR, and blue are not significant at
  10732. that threshold.
  10733. The alpha and beta globin genes targeted for blocking are marked with large
  10734. triangles, while all other genes are represented as small points.
  10735. \end_layout
  10736. \end_inset
  10737. \end_layout
  10738. \begin_layout Plain Layout
  10739. \end_layout
  10740. \end_inset
  10741. \end_layout
  10742. \begin_layout Standard
  10743. \begin_inset Flex TODO Note (inline)
  10744. status open
  10745. \begin_layout Plain Layout
  10746. Standardize on
  10747. \begin_inset Quotes eld
  10748. \end_inset
  10749. log2
  10750. \begin_inset Quotes erd
  10751. \end_inset
  10752. notation
  10753. \end_layout
  10754. \end_inset
  10755. \end_layout
  10756. \begin_layout Standard
  10757. The data do indeed show small systematic perturbations in gene levels (Figure
  10758. \begin_inset CommandInset ref
  10759. LatexCommand ref
  10760. reference "fig:MA-plot"
  10761. plural "false"
  10762. caps "false"
  10763. noprefix "false"
  10764. \end_inset
  10765. ).
  10766. Other than the 3 designated alpha and beta globin genes, two other genes
  10767. stand out as having especially large negative log fold changes: HBD and
  10768. LOC1021365.
  10769. HBD, delta globin, is most likely targeted by the blocking oligos due to
  10770. high sequence homology with the other globin genes.
  10771. LOC1021365 is the aforementioned ncRNA that is reverse-complementary to
  10772. one of the alpha-like genes and that would be expected to be removed during
  10773. the globin blocking step.
  10774. All other genes appear in a cluster centered vertically at 0, and the vast
  10775. majority of genes in this cluster show an absolute log2(FC) of 0.5 or less.
  10776. Nevertheless, many of these small perturbations are still statistically
  10777. significant, indicating that the globin blocking oligos likely cause very
  10778. small but non-zero systematic perturbations in measured gene expression
  10779. levels.
  10780. \end_layout
  10781. \begin_layout Standard
  10782. \begin_inset Float figure
  10783. wide false
  10784. sideways false
  10785. status collapsed
  10786. \begin_layout Plain Layout
  10787. \align center
  10788. \begin_inset Graphics
  10789. filename graphics/Globin Paper/figure5 - corrplot.pdf
  10790. lyxscale 50
  10791. width 70col%
  10792. \end_inset
  10793. \end_layout
  10794. \begin_layout Plain Layout
  10795. \begin_inset Caption Standard
  10796. \begin_layout Plain Layout
  10797. \series bold
  10798. \begin_inset Argument 1
  10799. status collapsed
  10800. \begin_layout Plain Layout
  10801. Comparison of inter-sample gene abundance correlations with and without
  10802. globin blocking.
  10803. \end_layout
  10804. \end_inset
  10805. \begin_inset CommandInset label
  10806. LatexCommand label
  10807. name "fig:gene-abundance-correlations"
  10808. \end_inset
  10809. Comparison of inter-sample gene abundance correlations with and without
  10810. globin blocking (GB).
  10811. \series default
  10812. All libraries were normalized together as described in Figure 2, and genes
  10813. with an average abundance (logCPM, log2 counts per million reads counted)
  10814. less than -1 were filtered out.
  10815. Each gene’s logCPM was computed in each library using the edgeR cpm function.
  10816. For each pair of biological samples, the Pearson correlation between those
  10817. samples' GB libraries was plotted against the correlation between the same
  10818. samples’ non-GB libraries.
  10819. Each point represents an unique pair of samples.
  10820. The solid gray line shows a quantile-quantile plot of distribution of GB
  10821. correlations vs.
  10822. that of non-GB correlations.
  10823. The thin dashed line is the identity line, provided for reference.
  10824. \end_layout
  10825. \end_inset
  10826. \end_layout
  10827. \begin_layout Plain Layout
  10828. \end_layout
  10829. \end_inset
  10830. \end_layout
  10831. \begin_layout Standard
  10832. To evaluate the possibility of globin blocking causing random perturbations
  10833. and reducing sample quality, we computed the Pearson correlation between
  10834. logCPM values for every pair of samples with and without GB and plotted
  10835. them against each other (Figure
  10836. \begin_inset CommandInset ref
  10837. LatexCommand ref
  10838. reference "fig:gene-abundance-correlations"
  10839. plural "false"
  10840. caps "false"
  10841. noprefix "false"
  10842. \end_inset
  10843. ).
  10844. The plot indicated that the GB libraries have higher sample-to-sample correlati
  10845. ons than the non-GB libraries.
  10846. Parametric and nonparametric tests for differences between the correlations
  10847. with and without GB both confirmed that this difference was highly significant
  10848. (2-sided paired t-test: t = 37.2, df = 665, P ≪ 2.2e-16; 2-sided Wilcoxon
  10849. sign-rank test: V = 2195, P ≪ 2.2e-16).
  10850. Performing the same tests on the Spearman correlations gave the same conclusion
  10851. (t-test: t = 26.8, df = 665, P ≪ 2.2e-16; sign-rank test: V = 8781, P ≪ 2.2e-16).
  10852. The edgeR package was used to compute the overall biological coefficient
  10853. of variation (BCV) for GB and non-GB libraries, and found that globin blocking
  10854. resulted in a negligible increase in the BCV (0.417 with GB vs.
  10855. 0.400 without).
  10856. The near equality of the BCVs for both sets indicates that the higher correlati
  10857. ons in the GB libraries are most likely a result of the increased yield
  10858. of useful reads, which reduces the contribution of Poisson counting uncertainty
  10859. to the overall variance of the logCPM values
  10860. \begin_inset CommandInset citation
  10861. LatexCommand cite
  10862. key "McCarthy2012"
  10863. literal "false"
  10864. \end_inset
  10865. .
  10866. This improves the precision of expression measurements and more than offsets
  10867. the negligible increase in BCV.
  10868. \end_layout
  10869. \begin_layout Subsection
  10870. More differentially expressed genes are detected with globin blocking
  10871. \end_layout
  10872. \begin_layout Standard
  10873. \begin_inset Float table
  10874. wide false
  10875. sideways false
  10876. status collapsed
  10877. \begin_layout Plain Layout
  10878. \align center
  10879. \begin_inset Tabular
  10880. <lyxtabular version="3" rows="5" columns="5">
  10881. <features tabularvalignment="middle">
  10882. <column alignment="center" valignment="top">
  10883. <column alignment="center" valignment="top">
  10884. <column alignment="center" valignment="top">
  10885. <column alignment="center" valignment="top">
  10886. <column alignment="center" valignment="top">
  10887. <row>
  10888. <cell alignment="center" valignment="top" usebox="none">
  10889. \begin_inset Text
  10890. \begin_layout Plain Layout
  10891. \end_layout
  10892. \end_inset
  10893. </cell>
  10894. <cell alignment="center" valignment="top" usebox="none">
  10895. \begin_inset Text
  10896. \begin_layout Plain Layout
  10897. \end_layout
  10898. \end_inset
  10899. </cell>
  10900. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  10901. \begin_inset Text
  10902. \begin_layout Plain Layout
  10903. \series bold
  10904. No Globin Blocking
  10905. \end_layout
  10906. \end_inset
  10907. </cell>
  10908. <cell multicolumn="2" alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  10909. \begin_inset Text
  10910. \begin_layout Plain Layout
  10911. \end_layout
  10912. \end_inset
  10913. </cell>
  10914. <cell multicolumn="2" alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  10915. \begin_inset Text
  10916. \begin_layout Plain Layout
  10917. \end_layout
  10918. \end_inset
  10919. </cell>
  10920. </row>
  10921. <row>
  10922. <cell alignment="center" valignment="top" usebox="none">
  10923. \begin_inset Text
  10924. \begin_layout Plain Layout
  10925. \end_layout
  10926. \end_inset
  10927. </cell>
  10928. <cell alignment="center" valignment="top" usebox="none">
  10929. \begin_inset Text
  10930. \begin_layout Plain Layout
  10931. \end_layout
  10932. \end_inset
  10933. </cell>
  10934. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10935. \begin_inset Text
  10936. \begin_layout Plain Layout
  10937. \series bold
  10938. Up
  10939. \end_layout
  10940. \end_inset
  10941. </cell>
  10942. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10943. \begin_inset Text
  10944. \begin_layout Plain Layout
  10945. \series bold
  10946. NS
  10947. \end_layout
  10948. \end_inset
  10949. </cell>
  10950. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  10951. \begin_inset Text
  10952. \begin_layout Plain Layout
  10953. \series bold
  10954. Down
  10955. \end_layout
  10956. \end_inset
  10957. </cell>
  10958. </row>
  10959. <row>
  10960. <cell multirow="3" alignment="center" valignment="middle" topline="true" bottomline="true" leftline="true" usebox="none">
  10961. \begin_inset Text
  10962. \begin_layout Plain Layout
  10963. \series bold
  10964. Globin-Blocking
  10965. \end_layout
  10966. \end_inset
  10967. </cell>
  10968. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10969. \begin_inset Text
  10970. \begin_layout Plain Layout
  10971. \series bold
  10972. Up
  10973. \end_layout
  10974. \end_inset
  10975. </cell>
  10976. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10977. \begin_inset Text
  10978. \begin_layout Plain Layout
  10979. \family roman
  10980. \series medium
  10981. \shape up
  10982. \size normal
  10983. \emph off
  10984. \bar no
  10985. \strikeout off
  10986. \xout off
  10987. \uuline off
  10988. \uwave off
  10989. \noun off
  10990. \color none
  10991. 231
  10992. \end_layout
  10993. \end_inset
  10994. </cell>
  10995. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  10996. \begin_inset Text
  10997. \begin_layout Plain Layout
  10998. \family roman
  10999. \series medium
  11000. \shape up
  11001. \size normal
  11002. \emph off
  11003. \bar no
  11004. \strikeout off
  11005. \xout off
  11006. \uuline off
  11007. \uwave off
  11008. \noun off
  11009. \color none
  11010. 515
  11011. \end_layout
  11012. \end_inset
  11013. </cell>
  11014. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  11015. \begin_inset Text
  11016. \begin_layout Plain Layout
  11017. \family roman
  11018. \series medium
  11019. \shape up
  11020. \size normal
  11021. \emph off
  11022. \bar no
  11023. \strikeout off
  11024. \xout off
  11025. \uuline off
  11026. \uwave off
  11027. \noun off
  11028. \color none
  11029. 2
  11030. \end_layout
  11031. \end_inset
  11032. </cell>
  11033. </row>
  11034. <row>
  11035. <cell multirow="4" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  11036. \begin_inset Text
  11037. \begin_layout Plain Layout
  11038. \end_layout
  11039. \end_inset
  11040. </cell>
  11041. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  11042. \begin_inset Text
  11043. \begin_layout Plain Layout
  11044. \series bold
  11045. NS
  11046. \end_layout
  11047. \end_inset
  11048. </cell>
  11049. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  11050. \begin_inset Text
  11051. \begin_layout Plain Layout
  11052. \family roman
  11053. \series medium
  11054. \shape up
  11055. \size normal
  11056. \emph off
  11057. \bar no
  11058. \strikeout off
  11059. \xout off
  11060. \uuline off
  11061. \uwave off
  11062. \noun off
  11063. \color none
  11064. 160
  11065. \end_layout
  11066. \end_inset
  11067. </cell>
  11068. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  11069. \begin_inset Text
  11070. \begin_layout Plain Layout
  11071. \family roman
  11072. \series medium
  11073. \shape up
  11074. \size normal
  11075. \emph off
  11076. \bar no
  11077. \strikeout off
  11078. \xout off
  11079. \uuline off
  11080. \uwave off
  11081. \noun off
  11082. \color none
  11083. 11235
  11084. \end_layout
  11085. \end_inset
  11086. </cell>
  11087. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  11088. \begin_inset Text
  11089. \begin_layout Plain Layout
  11090. \family roman
  11091. \series medium
  11092. \shape up
  11093. \size normal
  11094. \emph off
  11095. \bar no
  11096. \strikeout off
  11097. \xout off
  11098. \uuline off
  11099. \uwave off
  11100. \noun off
  11101. \color none
  11102. 136
  11103. \end_layout
  11104. \end_inset
  11105. </cell>
  11106. </row>
  11107. <row>
  11108. <cell multirow="4" alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  11109. \begin_inset Text
  11110. \begin_layout Plain Layout
  11111. \end_layout
  11112. \end_inset
  11113. </cell>
  11114. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  11115. \begin_inset Text
  11116. \begin_layout Plain Layout
  11117. \series bold
  11118. Down
  11119. \end_layout
  11120. \end_inset
  11121. </cell>
  11122. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  11123. \begin_inset Text
  11124. \begin_layout Plain Layout
  11125. \family roman
  11126. \series medium
  11127. \shape up
  11128. \size normal
  11129. \emph off
  11130. \bar no
  11131. \strikeout off
  11132. \xout off
  11133. \uuline off
  11134. \uwave off
  11135. \noun off
  11136. \color none
  11137. 0
  11138. \end_layout
  11139. \end_inset
  11140. </cell>
  11141. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  11142. \begin_inset Text
  11143. \begin_layout Plain Layout
  11144. \family roman
  11145. \series medium
  11146. \shape up
  11147. \size normal
  11148. \emph off
  11149. \bar no
  11150. \strikeout off
  11151. \xout off
  11152. \uuline off
  11153. \uwave off
  11154. \noun off
  11155. \color none
  11156. 548
  11157. \end_layout
  11158. \end_inset
  11159. </cell>
  11160. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  11161. \begin_inset Text
  11162. \begin_layout Plain Layout
  11163. \family roman
  11164. \series medium
  11165. \shape up
  11166. \size normal
  11167. \emph off
  11168. \bar no
  11169. \strikeout off
  11170. \xout off
  11171. \uuline off
  11172. \uwave off
  11173. \noun off
  11174. \color none
  11175. 127
  11176. \end_layout
  11177. \end_inset
  11178. </cell>
  11179. </row>
  11180. </lyxtabular>
  11181. \end_inset
  11182. \end_layout
  11183. \begin_layout Plain Layout
  11184. \begin_inset Caption Standard
  11185. \begin_layout Plain Layout
  11186. \series bold
  11187. \begin_inset Argument 1
  11188. status open
  11189. \begin_layout Plain Layout
  11190. Comparison of significantly differentially expressed genes with and without
  11191. globin blocking.
  11192. \end_layout
  11193. \end_inset
  11194. \begin_inset CommandInset label
  11195. LatexCommand label
  11196. name "tab:Comparison-of-significant"
  11197. \end_inset
  11198. Comparison of significantly differentially expressed genes with and without
  11199. globin blocking.
  11200. \series default
  11201. Up, Down: Genes significantly up/down-regulated in post-transplant samples
  11202. relative to pre-transplant samples, with a false discovery rate of 10%
  11203. or less.
  11204. NS: Non-significant genes (false discovery rate greater than 10%).
  11205. \end_layout
  11206. \end_inset
  11207. \end_layout
  11208. \begin_layout Plain Layout
  11209. \end_layout
  11210. \end_inset
  11211. \end_layout
  11212. \begin_layout Standard
  11213. To compare performance on differential gene expression tests, we took subsets
  11214. of both the GB and non-GB libraries with exactly one pre-transplant and
  11215. one post-transplant sample for each animal that had paired samples available
  11216. for analysis (N=7 animals, N=14 samples in each subset).
  11217. The same test for pre- vs.
  11218. post-transplant differential gene expression was performed on the same
  11219. 7 pairs of samples from GB libraries and non-GB libraries, in each case
  11220. using an FDR of 10% as the threshold of significance.
  11221. Out of 12954 genes that passed the detection threshold in both subsets,
  11222. 358 were called significantly differentially expressed in the same direction
  11223. in both sets; 1063 were differentially expressed in the GB set only; 296
  11224. were differentially expressed in the non-GB set only; 2 genes were called
  11225. significantly up in the GB set but significantly down in the non-GB set;
  11226. and the remaining 11235 were not called differentially expressed in either
  11227. set.
  11228. These data are summarized in Table
  11229. \begin_inset CommandInset ref
  11230. LatexCommand ref
  11231. reference "tab:Comparison-of-significant"
  11232. plural "false"
  11233. caps "false"
  11234. noprefix "false"
  11235. \end_inset
  11236. .
  11237. The differences in BCV calculated by EdgeR for these subsets of samples
  11238. were negligible (BCV = 0.302 for GB and 0.297 for non-GB).
  11239. \end_layout
  11240. \begin_layout Standard
  11241. The key point is that the GB data results in substantially more differentially
  11242. expressed calls than the non-GB data.
  11243. Since there is no gold standard for this dataset, it is impossible to be
  11244. certain whether this is due to under-calling of differential expression
  11245. in the non-GB samples or over-calling in the GB samples.
  11246. However, given that both datasets are derived from the same biological
  11247. samples and have nearly equal BCVs, it is more likely that the larger number
  11248. of DE calls in the GB samples are genuine detections that were enabled
  11249. by the higher sequencing depth and measurement precision of the GB samples.
  11250. Note that the same set of genes was considered in both subsets, so the
  11251. larger number of differentially expressed gene calls in the GB data set
  11252. reflects a greater sensitivity to detect significant differential gene
  11253. expression and not simply the larger total number of detected genes in
  11254. GB samples described earlier.
  11255. \end_layout
  11256. \begin_layout Section
  11257. Discussion
  11258. \end_layout
  11259. \begin_layout Standard
  11260. The original experience with whole blood gene expression profiling on DNA
  11261. microarrays demonstrated that the high concentration of globin transcripts
  11262. reduced the sensitivity to detect genes with relatively low expression
  11263. levels, in effect, significantly reducing the sensitivity.
  11264. To address this limitation, commercial protocols for globin reduction were
  11265. developed based on strategies to block globin transcript amplification
  11266. during labeling or physically removing globin transcripts by affinity bead
  11267. methods
  11268. \begin_inset CommandInset citation
  11269. LatexCommand cite
  11270. key "Winn2010"
  11271. literal "false"
  11272. \end_inset
  11273. .
  11274. More recently, using the latest generation of labeling protocols and arrays,
  11275. it was determined that globin reduction was no longer necessary to obtain
  11276. sufficient sensitivity to detect differential transcript expression
  11277. \begin_inset CommandInset citation
  11278. LatexCommand cite
  11279. key "NuGEN2010"
  11280. literal "false"
  11281. \end_inset
  11282. .
  11283. However, we are not aware of any publications using these currently available
  11284. protocols the with latest generation of microarrays that actually compare
  11285. the detection sensitivity with and without globin reduction.
  11286. However, in practice this has now been adopted generally primarily driven
  11287. by concerns for cost control.
  11288. The main objective of our work was to directly test the impact of globin
  11289. gene transcripts and a new globin blocking protocol for application to
  11290. the newest generation of differential gene expression profiling determined
  11291. using next generation sequencing.
  11292. \end_layout
  11293. \begin_layout Standard
  11294. The challenge of doing global gene expression profiling in cynomolgus monkeys
  11295. is that the current available arrays were never designed to comprehensively
  11296. cover this genome and have not been updated since the first assemblies
  11297. of the cynomolgus genome were published.
  11298. Therefore, we determined that the best strategy for peripheral blood profiling
  11299. was to do deep RNA-seq and inform the workflow using the latest available
  11300. genome assembly and annotation
  11301. \begin_inset CommandInset citation
  11302. LatexCommand cite
  11303. key "Wilson2013"
  11304. literal "false"
  11305. \end_inset
  11306. .
  11307. However, it was not immediately clear whether globin reduction was necessary
  11308. for RNA-seq or how much improvement in efficiency or sensitivity to detect
  11309. differential gene expression would be achieved for the added cost and work.
  11310. \end_layout
  11311. \begin_layout Standard
  11312. We only found one report that demonstrated that globin reduction significantly
  11313. improved the effective read yields for sequencing of human peripheral blood
  11314. cell RNA using a DeepSAGE protocol
  11315. \begin_inset CommandInset citation
  11316. LatexCommand cite
  11317. key "Mastrokolias2012"
  11318. literal "false"
  11319. \end_inset
  11320. .
  11321. The approach to DeepSAGE involves two different restriction enzymes that
  11322. purify and then tag small fragments of transcripts at specific locations
  11323. and thus, significantly reduces the complexity of the transcriptome.
  11324. Therefore, we could not determine how DeepSAGE results would translate
  11325. to the common strategy in the field for assaying the entire transcript
  11326. population by whole-transcriptome 3’-end RNA-seq.
  11327. Furthermore, if globin reduction is necessary, we also needed a globin
  11328. reduction method specific to cynomolgus globin sequences that would work
  11329. an organism for which no kit is available off the shelf.
  11330. \end_layout
  11331. \begin_layout Standard
  11332. As mentioned above, the addition of globin blocking oligos has a very small
  11333. impact on measured expression levels of gene expression.
  11334. However, this is a non-issue for the purposes of differential expression
  11335. testing, since a systematic change in a gene in all samples does not affect
  11336. relative expression levels between samples.
  11337. However, we must acknowledge that simple comparisons of gene expression
  11338. data obtained by GB and non-GB protocols are not possible without additional
  11339. normalization.
  11340. \end_layout
  11341. \begin_layout Standard
  11342. More importantly, globin blocking not only nearly doubles the yield of usable
  11343. reads, it also increases inter-sample correlation and sensitivity to detect
  11344. differential gene expression relative to the same set of samples profiled
  11345. without blocking.
  11346. In addition, globin blocking does not add a significant amount of random
  11347. noise to the data.
  11348. Globin blocking thus represents a cost-effective way to squeeze more data
  11349. and statistical power out of the same blood samples and the same amount
  11350. of sequencing.
  11351. In conclusion, globin reduction greatly increases the yield of useful RNA-seq
  11352. reads mapping to the rest of the genome, with minimal perturbations in
  11353. the relative levels of non-globin genes.
  11354. Based on these results, globin transcript reduction using sequence-specific,
  11355. complementary blocking oligonucleotides is recommended for all deep RNA-seq
  11356. of cynomolgus and other nonhuman primate blood samples.
  11357. \end_layout
  11358. \begin_layout Section
  11359. Future Directions
  11360. \end_layout
  11361. \begin_layout Standard
  11362. One drawback of the globin blocking method presented in this analysis is
  11363. a poor yield of genic reads, only around 50%.
  11364. In a separate experiment, the reagent mixture was modified so as to address
  11365. this drawback, resulting in a method that produces an even better reduction
  11366. in globin reads without reducing the overall fraction of genic reads.
  11367. However, the data showing this improvement consists of only a few test
  11368. samples, so the larger data set analyzed above was chosen in order to demonstra
  11369. te the effectiveness of the method in reducing globin reads while preserving
  11370. the biological signal.
  11371. \end_layout
  11372. \begin_layout Standard
  11373. The motivation for developing a fast practical way to enrich for non-globin
  11374. reads in cyno blood samples was to enable a large-scale RNA-seq experiment
  11375. investigating the effects of mesenchymal stem cell infusion on blood gene
  11376. expression in cynomologus transplant recipients in a time course after
  11377. transplantation.
  11378. With the globin blocking method in place, the way is now clear for this
  11379. experiment to proceed.
  11380. \end_layout
  11381. \begin_layout Chapter
  11382. Future Directions
  11383. \end_layout
  11384. \begin_layout Standard
  11385. \begin_inset Flex TODO Note (inline)
  11386. status open
  11387. \begin_layout Plain Layout
  11388. If there are any chapter-independent future directions, put them here.
  11389. Otherwise, delete this section.
  11390. Check in the directions if this is OK.
  11391. \end_layout
  11392. \end_inset
  11393. \end_layout
  11394. \begin_layout Chapter
  11395. Closing remarks
  11396. \end_layout
  11397. \begin_layout Standard
  11398. \begin_inset ERT
  11399. status collapsed
  11400. \begin_layout Plain Layout
  11401. % Use "References" as the title of the Bibliography
  11402. \end_layout
  11403. \begin_layout Plain Layout
  11404. \backslash
  11405. renewcommand{
  11406. \backslash
  11407. bibname}{References}
  11408. \end_layout
  11409. \end_inset
  11410. \end_layout
  11411. \begin_layout Standard
  11412. \begin_inset CommandInset bibtex
  11413. LatexCommand bibtex
  11414. btprint "btPrintCited"
  11415. bibfiles "code-refs,refs-PROCESSED"
  11416. options "bibtotoc,unsrt"
  11417. \end_inset
  11418. \end_layout
  11419. \begin_layout Standard
  11420. \begin_inset Flex TODO Note (inline)
  11421. status open
  11422. \begin_layout Plain Layout
  11423. Check bib entry formatting & sort order
  11424. \end_layout
  11425. \end_inset
  11426. \end_layout
  11427. \begin_layout Standard
  11428. \begin_inset Flex TODO Note (inline)
  11429. status open
  11430. \begin_layout Plain Layout
  11431. Check in-text citation format.
  11432. Probably don't just want [1], [2], etc.
  11433. \end_layout
  11434. \end_inset
  11435. \end_layout
  11436. \end_body
  11437. \end_document