thesis.lyx 125 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058405940604061406240634064406540664067406840694070407140724073407440754076407740784079408040814082408340844085408640874088408940904091409240934094409540964097409840994100410141024103410441054106410741084109411041114112411341144115411641174118411941204121412241234124412541264127412841294130413141324133413441354136413741384139414041414142414341444145414641474148414941504151415241534154415541564157415841594160416141624163416441654166416741684169417041714172417341744175417641774178417941804181418241834184418541864187418841894190419141924193419441954196419741984199420042014202420342044205420642074208420942104211421242134214421542164217421842194220422142224223422442254226422742284229423042314232423342344235423642374238423942404241424242434244424542464247424842494250425142524253425442554256425742584259426042614262426342644265426642674268426942704271427242734274427542764277427842794280428142824283428442854286428742884289429042914292429342944295429642974298429943004301430243034304430543064307430843094310431143124313431443154316431743184319432043214322432343244325432643274328432943304331433243334334433543364337433843394340434143424343434443454346434743484349435043514352435343544355435643574358435943604361436243634364436543664367436843694370437143724373437443754376437743784379438043814382438343844385438643874388438943904391439243934394439543964397439843994400440144024403440444054406440744084409441044114412441344144415441644174418441944204421442244234424442544264427442844294430443144324433443444354436443744384439444044414442444344444445444644474448444944504451445244534454445544564457445844594460446144624463446444654466446744684469447044714472447344744475447644774478447944804481448244834484448544864487448844894490449144924493449444954496449744984499450045014502450345044505450645074508450945104511451245134514451545164517451845194520452145224523452445254526452745284529453045314532453345344535453645374538453945404541454245434544454545464547454845494550455145524553455445554556455745584559456045614562456345644565456645674568456945704571457245734574457545764577457845794580458145824583458445854586458745884589459045914592459345944595459645974598459946004601460246034604460546064607460846094610461146124613461446154616461746184619462046214622462346244625462646274628462946304631463246334634463546364637463846394640464146424643464446454646464746484649465046514652465346544655465646574658465946604661466246634664466546664667466846694670467146724673467446754676467746784679468046814682468346844685468646874688468946904691469246934694469546964697469846994700470147024703470447054706470747084709471047114712471347144715471647174718471947204721472247234724472547264727472847294730473147324733473447354736473747384739474047414742474347444745474647474748474947504751475247534754475547564757475847594760476147624763476447654766476747684769477047714772477347744775477647774778477947804781478247834784478547864787478847894790479147924793479447954796479747984799480048014802480348044805480648074808480948104811481248134814481548164817481848194820482148224823482448254826482748284829483048314832483348344835483648374838483948404841484248434844484548464847484848494850485148524853485448554856485748584859486048614862486348644865486648674868486948704871487248734874487548764877487848794880488148824883488448854886488748884889489048914892489348944895489648974898489949004901490249034904490549064907490849094910491149124913491449154916491749184919492049214922492349244925492649274928492949304931493249334934493549364937493849394940494149424943494449454946494749484949495049514952495349544955495649574958495949604961496249634964496549664967496849694970497149724973497449754976497749784979498049814982498349844985498649874988498949904991499249934994499549964997499849995000500150025003500450055006500750085009501050115012501350145015501650175018501950205021502250235024502550265027502850295030503150325033503450355036503750385039504050415042504350445045504650475048504950505051505250535054505550565057505850595060506150625063506450655066506750685069507050715072507350745075507650775078507950805081508250835084508550865087508850895090509150925093509450955096509750985099510051015102510351045105510651075108510951105111511251135114511551165117511851195120512151225123512451255126512751285129513051315132513351345135513651375138513951405141514251435144514551465147514851495150515151525153515451555156515751585159516051615162516351645165516651675168516951705171517251735174517551765177517851795180518151825183518451855186518751885189519051915192519351945195519651975198519952005201520252035204520552065207520852095210521152125213521452155216521752185219522052215222522352245225522652275228522952305231523252335234523552365237523852395240524152425243524452455246524752485249525052515252525352545255525652575258525952605261526252635264526552665267526852695270527152725273527452755276527752785279528052815282528352845285528652875288528952905291529252935294529552965297529852995300530153025303530453055306530753085309531053115312531353145315531653175318531953205321532253235324532553265327532853295330533153325333533453355336533753385339534053415342534353445345534653475348534953505351535253535354535553565357535853595360536153625363536453655366536753685369537053715372537353745375537653775378537953805381538253835384538553865387538853895390539153925393
  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. \SetWatermarkLightness{0.97}
  14. \SetWatermarkScale{1}
  15. % Set up required header format
  16. \usepackage{fancyhdr}
  17. \pagestyle{fancy}
  18. \renewcommand{\headrulewidth}{0pt}
  19. \rhead{}
  20. \lhead{}
  21. \rfoot{}
  22. \lfoot{}
  23. \cfoot{\thepage} % Page number bottom center
  24. % https://tex.stackexchange.com/questions/65680/automatically-bold-first-sentence-of-a-floats-caption
  25. \usepackage{xstring}
  26. \usepackage{etoolbox}
  27. \usepackage{caption}
  28. \captionsetup{labelfont=bf,tableposition=top}
  29. \makeatletter
  30. \newcommand\formatlabel[1]{%
  31. \noexpandarg
  32. \IfSubStr{#1}{.}{%
  33. \StrBefore{#1}{.}[\firstcaption]%
  34. \StrBehind{#1}{.}[\secondcaption]%
  35. \textbf{\firstcaption.} \secondcaption}{%
  36. #1}%
  37. }
  38. \patchcmd{\@caption}{#3}{\formatlabel{#3}}
  39. \makeatother
  40. \end_preamble
  41. \use_default_options true
  42. \begin_modules
  43. todonotes
  44. \end_modules
  45. \maintain_unincluded_children false
  46. \language english
  47. \language_package default
  48. \inputencoding utf8
  49. \fontencoding default
  50. \font_roman "default" "default"
  51. \font_sans "default" "default"
  52. \font_typewriter "default" "default"
  53. \font_math "auto" "auto"
  54. \font_default_family default
  55. \use_non_tex_fonts false
  56. \font_sc false
  57. \font_osf false
  58. \font_sf_scale 100 100
  59. \font_tt_scale 100 100
  60. \use_microtype false
  61. \use_dash_ligatures true
  62. \graphics default
  63. \default_output_format pdf4
  64. \output_sync 0
  65. \bibtex_command default
  66. \index_command default
  67. \paperfontsize 12
  68. \spacing double
  69. \use_hyperref true
  70. \pdf_bookmarks true
  71. \pdf_bookmarksnumbered false
  72. \pdf_bookmarksopen false
  73. \pdf_bookmarksopenlevel 1
  74. \pdf_breaklinks false
  75. \pdf_pdfborder false
  76. \pdf_colorlinks false
  77. \pdf_backref false
  78. \pdf_pdfusetitle true
  79. \papersize letterpaper
  80. \use_geometry true
  81. \use_package amsmath 1
  82. \use_package amssymb 1
  83. \use_package cancel 1
  84. \use_package esint 1
  85. \use_package mathdots 1
  86. \use_package mathtools 1
  87. \use_package mhchem 1
  88. \use_package stackrel 1
  89. \use_package stmaryrd 1
  90. \use_package undertilde 1
  91. \cite_engine basic
  92. \cite_engine_type default
  93. \biblio_style plain
  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 2
  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. May 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 Standard
  194. \begin_inset Flex TODO Note (inline)
  195. status open
  196. \begin_layout Plain Layout
  197. Look into auto-generated nomenclature list: https://wiki.lyx.org/Tips/Nomenclature
  198. \end_layout
  199. \end_inset
  200. \end_layout
  201. \begin_layout List of TODOs
  202. \end_layout
  203. \begin_layout Standard
  204. [Abstract]
  205. \end_layout
  206. \begin_layout Chapter*
  207. Abstract
  208. \end_layout
  209. \begin_layout Chapter
  210. Introduction
  211. \end_layout
  212. \begin_layout Section
  213. Background & Significance
  214. \end_layout
  215. \begin_layout Subsection
  216. Biological motivation
  217. \end_layout
  218. \begin_layout Itemize
  219. Rejection is the major long-term threat to organ and tissue grafts
  220. \end_layout
  221. \begin_deeper
  222. \begin_layout Itemize
  223. Common mechanisms of rejection
  224. \end_layout
  225. \begin_layout Itemize
  226. Effective immune suppression requires monitoring for rejection and tuning
  227. \end_layout
  228. \begin_layout Itemize
  229. Current tests for rejection (tissue biopsy) are invasive and biased
  230. \end_layout
  231. \begin_layout Itemize
  232. A blood test based on microarrays would be less biased and invasive
  233. \end_layout
  234. \end_deeper
  235. \begin_layout Itemize
  236. Memory cells are resistant to immune suppression
  237. \end_layout
  238. \begin_deeper
  239. \begin_layout Itemize
  240. Mechanisms of resistance in memory cells are poorly understood
  241. \end_layout
  242. \begin_layout Itemize
  243. A better understanding of immune memory formation is needed
  244. \end_layout
  245. \end_deeper
  246. \begin_layout Itemize
  247. Mesenchymal stem cell infusion is a promising new treatment to prevent/delay
  248. rejection
  249. \end_layout
  250. \begin_deeper
  251. \begin_layout Itemize
  252. Demonstrated in mice, but not yet in primates
  253. \end_layout
  254. \begin_layout Itemize
  255. Mechanism currently unknown, but MSC are known to be immune modulatory
  256. \end_layout
  257. \end_deeper
  258. \begin_layout Subsection
  259. Overview of bioinformatic analysis methods
  260. \end_layout
  261. \begin_layout Standard
  262. An overview of all the methods used, including what problem they solve,
  263. what assumptions they make, and a basic description of how they work.
  264. \end_layout
  265. \begin_layout Itemize
  266. ChIP-seq Peak calling
  267. \end_layout
  268. \begin_deeper
  269. \begin_layout Itemize
  270. Cross-correlation analysis to determine fragment size
  271. \end_layout
  272. \begin_layout Itemize
  273. Broad vs narrow peaks
  274. \end_layout
  275. \begin_layout Itemize
  276. SICER for broad peaks
  277. \end_layout
  278. \begin_layout Itemize
  279. IDR for biologically reproducible peaks
  280. \end_layout
  281. \begin_layout Itemize
  282. csaw peak filtering guidelines for unbiased downstream analysis
  283. \end_layout
  284. \end_deeper
  285. \begin_layout Itemize
  286. Normalization is non-trivial and application-dependant
  287. \end_layout
  288. \begin_deeper
  289. \begin_layout Itemize
  290. Expression arrays: RMA & fRMA; why fRMA is needed
  291. \end_layout
  292. \begin_layout Itemize
  293. Methylation arrays: M-value transformation approximates normal data but
  294. induces heteroskedasticity
  295. \end_layout
  296. \begin_layout Itemize
  297. RNA-seq: normalize based on assumption that the average gene is not changing
  298. \end_layout
  299. \begin_layout Itemize
  300. ChIP-seq: complex with many considerations, dependent on experimental methods,
  301. biological system, and analysis goals
  302. \end_layout
  303. \end_deeper
  304. \begin_layout Itemize
  305. Limma: The standard linear modeling framework for genomics
  306. \end_layout
  307. \begin_deeper
  308. \begin_layout Itemize
  309. empirical Bayes variance modeling: limma's core feature
  310. \end_layout
  311. \begin_layout Itemize
  312. edgeR & DESeq2: Extend with negative bonomial GLM for RNA-seq and other
  313. count data
  314. \end_layout
  315. \begin_layout Itemize
  316. voom: Extend with precision weights to model mean-variance trend
  317. \end_layout
  318. \begin_layout Itemize
  319. arrayWeights and duplicateCorrelation to handle complex variance structures
  320. \end_layout
  321. \end_deeper
  322. \begin_layout Itemize
  323. sva and ComBat for batch correction
  324. \end_layout
  325. \begin_layout Itemize
  326. Factor analysis: PCA, MDS, MOFA
  327. \end_layout
  328. \begin_deeper
  329. \begin_layout Itemize
  330. Batch-corrected PCA is informative, but careful application is required
  331. to avoid bias
  332. \end_layout
  333. \end_deeper
  334. \begin_layout Itemize
  335. Gene set analysis: camera and SPIA
  336. \end_layout
  337. \begin_layout Section
  338. Innovation
  339. \end_layout
  340. \begin_layout Itemize
  341. MSC infusion to improve transplant outcomes (prevent/delay rejection)
  342. \end_layout
  343. \begin_deeper
  344. \begin_layout Itemize
  345. Characterize MSC response to interferon gamma
  346. \end_layout
  347. \begin_layout Itemize
  348. IFN-g is thought to stimulate their function
  349. \end_layout
  350. \begin_layout Itemize
  351. Test IFN-g treated MSC infusion as a therapy to delay graft rejection in
  352. cynomolgus monkeys
  353. \end_layout
  354. \begin_layout Itemize
  355. Monitor animals post-transplant using blood RNA-seq at serial time points
  356. \end_layout
  357. \end_deeper
  358. \begin_layout Itemize
  359. Investigate dynamics of histone marks in CD4 T-cell activation and memory
  360. \end_layout
  361. \begin_deeper
  362. \begin_layout Itemize
  363. Previous studies have looked at single snapshots of histone marks
  364. \end_layout
  365. \begin_layout Itemize
  366. Instead, look at changes in histone marks across activation and memory
  367. \end_layout
  368. \end_deeper
  369. \begin_layout Itemize
  370. High-throughput sequencing and microarray technologies
  371. \end_layout
  372. \begin_deeper
  373. \begin_layout Itemize
  374. Powerful methods for assaying gene expression and epigenetics across entire
  375. genomes
  376. \end_layout
  377. \begin_layout Itemize
  378. Proper analysis requires finding and exploiting systematic genome-wide trends
  379. \end_layout
  380. \end_deeper
  381. \begin_layout Chapter
  382. Reproducible genome-wide epigenetic analysis of H3K4 and H3K27 methylation
  383. in naive and memory CD4 T-cell activation
  384. \end_layout
  385. \begin_layout Standard
  386. \begin_inset Flex TODO Note (inline)
  387. status open
  388. \begin_layout Plain Layout
  389. Author list: Me, Sarah, Dan
  390. \end_layout
  391. \end_inset
  392. \end_layout
  393. \begin_layout Section
  394. Approach
  395. \end_layout
  396. \begin_layout Itemize
  397. CD4 T-cells are central to all adaptive immune responses and memory
  398. \end_layout
  399. \begin_layout Itemize
  400. H3K4 and H3K27 methylation are major epigenetic regulators of gene expression
  401. \end_layout
  402. \begin_layout Itemize
  403. Canonically, H3K4 is activating and H3K27 is inhibitory, but the reality
  404. is complex
  405. \end_layout
  406. \begin_layout Itemize
  407. Looking at these marks during CD4 activation and memory should reveal new
  408. mechanistic details
  409. \end_layout
  410. \begin_layout Itemize
  411. Test
  412. \begin_inset Quotes eld
  413. \end_inset
  414. poised promoter
  415. \begin_inset Quotes erd
  416. \end_inset
  417. hypothesis in which H3K4 and H3K27 are both methylated
  418. \end_layout
  419. \begin_layout Itemize
  420. Expand scope of analysis beyond simple promoter counts
  421. \end_layout
  422. \begin_deeper
  423. \begin_layout Itemize
  424. Analyze peaks genome-wide, including in intergenic regions
  425. \end_layout
  426. \begin_layout Itemize
  427. Analysis of coverage distribution shape within promoters, e.g.
  428. upstream vs downstream coverage
  429. \end_layout
  430. \end_deeper
  431. \begin_layout Section
  432. Methods
  433. \end_layout
  434. \begin_layout Itemize
  435. Re-analyze previously published CD4 ChIP-seq & RNA-seq data
  436. \begin_inset CommandInset citation
  437. LatexCommand cite
  438. key "LaMere2016,Lamere2017"
  439. literal "true"
  440. \end_inset
  441. \end_layout
  442. \begin_deeper
  443. \begin_layout Itemize
  444. Completely reimplement analysis from scratch as a reproducible workflow
  445. \end_layout
  446. \begin_layout Itemize
  447. Use newly published methods & algorithms not available during the original
  448. analysis: SICER, csaw, MOFA, ComBat, sva, GREAT, and more
  449. \end_layout
  450. \end_deeper
  451. \begin_layout Itemize
  452. SICER, IDR, csaw, & GREAT to call ChIP-seq peaks genome-wide, perform differenti
  453. al abundance analysis, and relate those peaks to gene expression
  454. \end_layout
  455. \begin_layout Itemize
  456. Promoter counts in sliding windows around each gene's highest-expressed
  457. TSS to investigate coverage distribution within promoters
  458. \end_layout
  459. \begin_layout Section
  460. Results
  461. \end_layout
  462. \begin_layout Standard
  463. \begin_inset Note Note
  464. status open
  465. \begin_layout Plain Layout
  466. Focus on what hypotheses were tested, then select figures that show how
  467. those hypotheses were tested, even if the result is a negative.
  468. \end_layout
  469. \end_inset
  470. \end_layout
  471. \begin_layout Subsection
  472. H3K4 and H3K27 methylation occur in broad regions and are enriched near
  473. promoters
  474. \end_layout
  475. \begin_layout Itemize
  476. Figures comparing MACS (non-broad peak caller) to SICER/epic (broad peak
  477. caller)
  478. \end_layout
  479. \begin_deeper
  480. \begin_layout Itemize
  481. Compare peak sizes and number of called peaks
  482. \end_layout
  483. \begin_layout Itemize
  484. Show representative IDR consistency plots for both
  485. \end_layout
  486. \end_deeper
  487. \begin_layout Itemize
  488. IDR analysis shows that SICER-called peaks are much more reproducible between
  489. biological replicates
  490. \end_layout
  491. \begin_layout Itemize
  492. Each histone mark is enriched within a certain radius of gene TSS positions,
  493. but that radius is different for each mark (figure)
  494. \end_layout
  495. \begin_layout Subsection
  496. RNA-seq has a large confounding batch effect
  497. \end_layout
  498. \begin_layout Itemize
  499. RNA-seq batch effect can be partially corrected, but still induces uncorrectable
  500. biases in downstream analysis
  501. \end_layout
  502. \begin_deeper
  503. \begin_layout Itemize
  504. Figure showing MDS plot before & after ComBat
  505. \end_layout
  506. \begin_layout Itemize
  507. Figure relating sample weights to batches, cell types, time points, etc.,
  508. showing that one batch is significantly worse quality
  509. \end_layout
  510. \begin_layout Itemize
  511. Figures showing p-value histograms for within-batch and cross-batch contrasts,
  512. showing that cross-batch contrasts have attenuated signal, as do comparisons
  513. within the bad batch
  514. \end_layout
  515. \end_deeper
  516. \begin_layout Subsection
  517. ChIP-seq must be corrected for hidden confounding factors
  518. \end_layout
  519. \begin_layout Itemize
  520. Figures showing pre- and post-SVA MDS plots for each histone mark
  521. \end_layout
  522. \begin_layout Itemize
  523. Figures showing BCV plots with and without SVA for each histone mark
  524. \end_layout
  525. \begin_layout Subsection
  526. H3K4 and H3K27 promoter methylation has broadly the expected correlation
  527. with gene expression
  528. \end_layout
  529. \begin_layout Itemize
  530. H3K4 is correlated with higher expression, and H3K27 is correlated with
  531. lower expression genome-wide
  532. \end_layout
  533. \begin_layout Itemize
  534. Figures showing these correlations: box/violin plots of expression distributions
  535. with every combination of peak presence/absence in promoter
  536. \end_layout
  537. \begin_layout Itemize
  538. Appropriate statistical tests showing significant differences in expected
  539. directions
  540. \end_layout
  541. \begin_layout Subsection
  542. MOFA recovers biologically relevant variation from blind analysis by correlating
  543. across datasets
  544. \end_layout
  545. \begin_layout Itemize
  546. MOFA
  547. \begin_inset CommandInset citation
  548. LatexCommand cite
  549. key "Argelaguet2018"
  550. literal "false"
  551. \end_inset
  552. successfully separates biologically relevant patterns of variation from
  553. technical confounding factors without knowing the sample labels, by finding
  554. latent factors that explain variation across multiple data sets.
  555. \end_layout
  556. \begin_deeper
  557. \begin_layout Itemize
  558. Figure: show percent-variance-explained plot from MOFA and PCA-like plots
  559. for the relevant latent factors
  560. \end_layout
  561. \begin_layout Itemize
  562. MOFA analysis also shows that batch effect correction can't get much better
  563. than it already is (Figure comparing blind MOFA batch correction to ComBat
  564. correction)
  565. \end_layout
  566. \end_deeper
  567. \begin_layout Subsection
  568. Naive-to-memory convergence observed in H3K4 and RNA-seq data, not in H3K27me3
  569. \end_layout
  570. \begin_layout Itemize
  571. H3K4 and RNA-seq data show clear evidence of naive convergence with memory
  572. between days 1 and 5 (MDS plot figure, also compare with last figure from
  573. \begin_inset CommandInset citation
  574. LatexCommand cite
  575. key "LaMere2016"
  576. literal "false"
  577. \end_inset
  578. )
  579. \end_layout
  580. \begin_layout Standard
  581. \begin_inset Flex TODO Note (inline)
  582. status open
  583. \begin_layout Plain Layout
  584. Get explicit permission from Sarah to include the figure
  585. \end_layout
  586. \end_inset
  587. \end_layout
  588. \begin_layout Itemize
  589. Table of numbers of genes different between N & M at each time point, showing
  590. dwindling differences at later time points, consistent with convergence
  591. \end_layout
  592. \begin_layout Itemize
  593. Similar figure for H3K27me3 showing lack of convergence
  594. \end_layout
  595. \begin_layout Subsection
  596. Effect of promoter coverage upstream vs downstream of TSS
  597. \end_layout
  598. \begin_layout Itemize
  599. H3K4me peaks seem to correlate with increased expression as long as they
  600. are anywhere near the TSS
  601. \end_layout
  602. \begin_layout Itemize
  603. H3K27me3 peaks can have different correlations to gene expression depending
  604. on their position relative to TSS (e.g.
  605. upstream vs downstream) Results consistent with
  606. \begin_inset CommandInset citation
  607. LatexCommand cite
  608. key "Young2011"
  609. literal "false"
  610. \end_inset
  611. \end_layout
  612. \begin_layout Section
  613. Discussion
  614. \end_layout
  615. \begin_layout Itemize
  616. "Promoter radius" is not constant and must be defined empirically for a
  617. given data set
  618. \end_layout
  619. \begin_layout Itemize
  620. MOFA shows great promise for accelerating discovery of major biological
  621. effects in multi-omics datasets
  622. \end_layout
  623. \begin_deeper
  624. \begin_layout Itemize
  625. MOFA was added to this analysis late and played primarily a confirmatory
  626. role, but it was able to confirm earlier conclusions with much less prior
  627. information (no sample labels) and much less analyst effort
  628. \end_layout
  629. \begin_layout Itemize
  630. MOFA confirmed that the already-implemented batch correction in the RNA-seq
  631. data was already performing as well as possible given the limitations of
  632. the data
  633. \end_layout
  634. \end_deeper
  635. \begin_layout Itemize
  636. Naive-to-memory convergence implies that naive cells are differentiating
  637. into memory cells, and that gene expression and H3K4 methylation are involved
  638. in this differentiation while H3K27me3 is less involved
  639. \end_layout
  640. \begin_layout Itemize
  641. H3K27me3, canonically regarded as a deactivating mark, seems to have a more
  642. complex
  643. \end_layout
  644. \begin_layout Itemize
  645. Discuss advantages of developing using a reproducible workflow
  646. \end_layout
  647. \begin_layout Chapter
  648. Improving array-based analyses of transplant rejection by optimizing data
  649. preprocessing
  650. \end_layout
  651. \begin_layout Standard
  652. \begin_inset Note Note
  653. status open
  654. \begin_layout Plain Layout
  655. Author list: Me, Sunil, Tom, Padma, Dan
  656. \end_layout
  657. \end_inset
  658. \end_layout
  659. \begin_layout Section
  660. Approach
  661. \end_layout
  662. \begin_layout Subsection
  663. Proper pre-processing is essential for array data
  664. \end_layout
  665. \begin_layout Standard
  666. \begin_inset Flex TODO Note (inline)
  667. status open
  668. \begin_layout Plain Layout
  669. This section could probably use some citations
  670. \end_layout
  671. \end_inset
  672. \end_layout
  673. \begin_layout Standard
  674. Microarrays, bead ararys, and similar assays produce raw data in the form
  675. of fluorescence intensity measurements, with the each intensity measurement
  676. proportional to the abundance of some fluorescently-labelled target DNA
  677. or RNA sequence that base pairs to a specific probe sequence.
  678. However, these measurements for each probe are also affected my many technical
  679. confounding factors, such as the concentration of target material, strength
  680. of off-target binding, and the sensitivity of the imaging sensor.
  681. Some array designs also use multiple probe sequences for each target.
  682. Hence, extensive pre-processing of array data is necessary to normalize
  683. out the effects of these technical factors and summarize the information
  684. from multiple probes to arrive at a single usable estimate of abundance
  685. or other relevant quantity, such as a ratio of two abundances, for each
  686. target.
  687. \end_layout
  688. \begin_layout Standard
  689. The choice of pre-processing algorithms used in the analysis of an array
  690. data set can have a large effect on the results of that analysis.
  691. However, despite their importance, these steps are often neglected or rushed
  692. in order to get to the more scientifically interesting analysis steps involving
  693. the actual biology of the system under study.
  694. Hence, it is often possible to achieve substantial gains in statistical
  695. power, model goodness-of-fit, or other relevant performance measures, by
  696. checking the assumptions made by each preprocessing step and choosing specific
  697. normalization methods tailored to the specific goals of the current analysis.
  698. \end_layout
  699. \begin_layout Subsection
  700. Normalization for clinical microarray classifiers must be single-channel
  701. \end_layout
  702. \begin_layout Subsubsection
  703. Standard normalization methods are unsuitable for clinical application
  704. \end_layout
  705. \begin_layout Standard
  706. As the cost of performing microarray assays falls, there is increasing interest
  707. in using genomic assays for diagnostic purposes, such as distinguishing
  708. healthy transplants (TX) from transplants undergoing acute rejection (AR)
  709. or acute dysfunction with no rejection (ADNR).
  710. However, the the standard normalization algorithm used for microarray data,
  711. Robust Multi-chip Average (RMA)
  712. \begin_inset CommandInset citation
  713. LatexCommand cite
  714. key "Irizarry2003a"
  715. literal "false"
  716. \end_inset
  717. , is not applicable in a clinical setting.
  718. Two of the steps in RMA, quantile normalization and probe summarization
  719. by median polish, depend on every array in the data set being normalized.
  720. This means that adding or removing any arrays from a data set changes the
  721. normalized values for all arrays, and data sets that have been normalized
  722. separately cannot be compared to each other.
  723. Hence, when using RMA, any arrays to be analyzed together must also be
  724. normalized together, and the set of arrays included in the data set must
  725. be held constant throughout an analysis.
  726. \end_layout
  727. \begin_layout Standard
  728. These limitations present serious impediments to the use of arrays as a
  729. diagnostic tool.
  730. When training a classifier, the samples to be classified must not be involved
  731. in any step of the training process, lest their inclusion bias the training
  732. process.
  733. Once a classifier is deployed in a clinical setting, the samples to be
  734. classified will not even
  735. \emph on
  736. exist
  737. \emph default
  738. at the time of training, so including them would be impossible even if
  739. it were statistically justifiable.
  740. Therefore, any machine learning application for microarrays demands that
  741. the normalized expression values computed for an array must depend only
  742. on information contained within that array.
  743. This would ensure that each array's normalization is independent of every
  744. other array, and that arrays normalized separately can still be compared
  745. to each other without bias.
  746. Such a normalization is commonly referred to as
  747. \begin_inset Quotes eld
  748. \end_inset
  749. single-channel normalization
  750. \begin_inset Quotes erd
  751. \end_inset
  752. .
  753. \end_layout
  754. \begin_layout Subsubsection
  755. Several strategies are available to meet clinical normalization requirements
  756. \end_layout
  757. \begin_layout Standard
  758. Frozen RMA (fRMA) addresses these concerns by replacing the quantile normalizati
  759. on and median polish with alternatives that do not introduce inter-array
  760. dependence, allowing each array to be normalized independently of all others
  761. \begin_inset CommandInset citation
  762. LatexCommand cite
  763. key "McCall2010"
  764. literal "false"
  765. \end_inset
  766. .
  767. Quantile normalization is performed against a pre-generated set of quantiles
  768. learned from a collection of 850 publically available arrays sampled from
  769. a wide variety of tissues in the Gene Expression Omnibus (GEO).
  770. Each array's probe intensity distribution is normalized against these pre-gener
  771. ated quantiles.
  772. The median polish step is replaced with a robust weighted average of probe
  773. intensities, using inverse variance weights learned from the same public
  774. GEO data.
  775. The result is a normalization that satisfies the requirements mentioned
  776. above: each array is normalized independently of all others, and any two
  777. normalized arrays can be compared directly to each other.
  778. \end_layout
  779. \begin_layout Standard
  780. One important limitation of fRMA is that it requires a separate reference
  781. data set from which to learn the parameters (reference quantiles and probe
  782. weights) that will be used to normalize each array.
  783. These parameters are specific to a given array platform, and pre-generated
  784. parameters are only provided for the most common platforms, such as Affymetrix
  785. hgu133plus2.
  786. For a less common platform, such as hthgu133pluspm, is is necessary to
  787. learn custom parameters from in-house data before fRMA can be used to normalize
  788. samples on that platform
  789. \begin_inset CommandInset citation
  790. LatexCommand cite
  791. key "McCall2011"
  792. literal "false"
  793. \end_inset
  794. .
  795. \end_layout
  796. \begin_layout Standard
  797. One other option is the aptly-named Single Channel Array Normalization (SCAN),
  798. which adapts a normalization method originally designed for tiling arrays
  799. \begin_inset CommandInset citation
  800. LatexCommand cite
  801. key "Piccolo2012"
  802. literal "false"
  803. \end_inset
  804. .
  805. SCAN is truly single-channel in that it does not require a set of normalization
  806. paramters estimated from an external set of reference samples like fRMA
  807. does.
  808. \end_layout
  809. \begin_layout Subsection
  810. Heteroskedasticity must be accounted for in methylation array data
  811. \end_layout
  812. \begin_layout Subsubsection
  813. Methylation array preprocessing induces heteroskedasticity
  814. \end_layout
  815. \begin_layout Standard
  816. DNA methylation arrays are a relatively new kind of assay that uses microarrays
  817. to measure the degree of methylation on cytosines in specific regions arrayed
  818. across the genome.
  819. First, bisulfite treatment converts all unmethylated cytosines to uracil
  820. (which then become thymine after amplication) while leaving methylated
  821. cytosines unaffected.
  822. Then, each target region is interrogated with two probes: one binds to
  823. the original genomic sequence and interrogates the level of methylated
  824. DNA, and the other binds to the sequence with all Cs replaced by Ts and
  825. interrogates the level of unmethylated DNA.
  826. \end_layout
  827. \begin_layout Standard
  828. \begin_inset Float figure
  829. wide false
  830. sideways false
  831. status collapsed
  832. \begin_layout Plain Layout
  833. \begin_inset Graphics
  834. filename graphics/methylvoom/sigmoid.pdf
  835. \end_inset
  836. \end_layout
  837. \begin_layout Plain Layout
  838. \begin_inset Caption Standard
  839. \begin_layout Plain Layout
  840. \begin_inset CommandInset label
  841. LatexCommand label
  842. name "fig:Sigmoid-beta-m-mapping"
  843. \end_inset
  844. \series bold
  845. Sigmoid shape of the mapping between β and M values
  846. \end_layout
  847. \end_inset
  848. \end_layout
  849. \end_inset
  850. \end_layout
  851. \begin_layout Standard
  852. After normalization, these two probe intensities are summarized in one of
  853. two ways, each with advantages and disadvantages.
  854. β
  855. \series bold
  856. \series default
  857. values, interpreted as fraction of DNA copies methylated, range from 0 to
  858. 1.
  859. β
  860. \series bold
  861. \series default
  862. values are conceptually easy to interpret, but the constrained range makes
  863. them unsuitable for linear modeling, and their error distributions are
  864. highly non-normal, which also frustrates linear modeling.
  865. M-values, interpreted as the log ratio of methylated to unmethylated copies,
  866. are computed by mapping the beta values from
  867. \begin_inset Formula $[0,1]$
  868. \end_inset
  869. onto
  870. \begin_inset Formula $(-\infty,+\infty)$
  871. \end_inset
  872. using a sigmoid curve (Figure
  873. \begin_inset CommandInset ref
  874. LatexCommand ref
  875. reference "fig:Sigmoid-beta-m-mapping"
  876. plural "false"
  877. caps "false"
  878. noprefix "false"
  879. \end_inset
  880. ).
  881. This transformation results in values with better statistical perperties:
  882. the unconstrained range is suitable for linear modeling, and the error
  883. distributions are more normal.
  884. Hence, most linear modeling and other statistical testing on methylation
  885. arrays is performed using M-values.
  886. \end_layout
  887. \begin_layout Standard
  888. However, the steep slope of the sigmoid transformation near 0 and 1 tends
  889. to over-exaggerate small differences in β values near those extremes, which
  890. in turn amplifies the error in those values, leading to a U-shaped trend
  891. in the mean-variance curve: extreme values have higher variances than values
  892. near the middle.
  893. This mean-variance dependency must be accounted for when fitting the linear
  894. model for differential methylation, or else the variance will be systematically
  895. overestimated for probes with moderate M-values and underestimated for
  896. probes with extreme M-values.
  897. \end_layout
  898. \begin_layout Subsubsection
  899. The voom method for RNA-seq data can model M-value heteroskedasticity
  900. \end_layout
  901. \begin_layout Standard
  902. RNA-seq read count data are also known to show heteroskedasticity, and the
  903. voom method was developed for modeling this heteroskedasticity by estimating
  904. the mean-variance trend in the data and using this trend to assign precision
  905. weights to each observation
  906. \begin_inset CommandInset citation
  907. LatexCommand cite
  908. key "Law2013"
  909. literal "false"
  910. \end_inset
  911. .
  912. While methylation array data are not derived from counts and have a very
  913. different mean-variance relationship from that of typical RNA-seq data,
  914. the voom method makes no specific assumptions on the shape of the mean-variance
  915. relationship - it only assumes that the relationship is smooth enough to
  916. model using a lowess curve.
  917. Hence, the method is sufficiently general to model the mean-variance relationsh
  918. ip in methylation array data.
  919. However, the standard implementation of voom assumes that the input is
  920. given in raw read counts, and it must be adapted to run on methylation
  921. M-values.
  922. \end_layout
  923. \begin_layout Standard
  924. \begin_inset Flex TODO Note (inline)
  925. status open
  926. \begin_layout Plain Layout
  927. Put code on Github and reference it
  928. \end_layout
  929. \end_inset
  930. \end_layout
  931. \begin_layout Section
  932. Methods
  933. \end_layout
  934. \begin_layout Subsection
  935. Evaluation of classifier performance with different normalization methods
  936. \end_layout
  937. \begin_layout Standard
  938. For testing different normalizations, a data set of 157 hgu133plus2 arrays
  939. was used, consisting of blood samples from kidney transplant patients whose
  940. grafts had been graded as TX, AR, or ADNR via biopsy and histology (46
  941. TX, 69 AR, 42 ADNR)
  942. \begin_inset CommandInset citation
  943. LatexCommand cite
  944. key "Kurian2014"
  945. literal "true"
  946. \end_inset
  947. .
  948. Additionally, an external validation set of 75 samples was gathered from
  949. public GEO data (37 TX, 38 AR, no ADNR).
  950. \end_layout
  951. \begin_layout Standard
  952. \begin_inset Flex TODO Note (inline)
  953. status collapsed
  954. \begin_layout Plain Layout
  955. Find appropriate GEO identifiers if possible.
  956. Kurian 2014 says GSE15296, but this seems to be different data.
  957. I also need to look up the GEO accession for the external validation set.
  958. \end_layout
  959. \end_inset
  960. \end_layout
  961. \begin_layout Standard
  962. To evaluate the effect of each normalization on classifier performance,
  963. the same classifier training and validation procedure was used after each
  964. normalization method.
  965. The PAM package was used to train a nearest shrunken centroid classifier
  966. on the training set and select the appropriate threshold for centroid shrinking.
  967. Then the trained classifier was used to predict the class probabilities
  968. of each validation sample.
  969. From these class probabilities, ROC curves and area-under-curve (AUC) values
  970. were generated
  971. \begin_inset CommandInset citation
  972. LatexCommand cite
  973. key "Turck2011"
  974. literal "false"
  975. \end_inset
  976. .
  977. Each normalization was tested on two different sets of training and validation
  978. samples.
  979. For internal validation, the 115 TX and AR arrays in the internal set were
  980. split at random into two equal sized sets, one for training and one for
  981. validation, each containing the same numbers of TX and AR samples as the
  982. other set.
  983. For external validation, the full set of 115 TX and AR samples were used
  984. as a training set, and the 75 external TX and AR samples were used as the
  985. validation set.
  986. Thus, 2 ROC curves and AUC values were generated for each normalization
  987. method: one internal and one external.
  988. Because the external validation set contains no ADNR samples, only classificati
  989. on of TX and AR samples was considered.
  990. The ADNR samples were included during normalization but excluded from all
  991. classifier training and validation.
  992. This ensures that the performance on internal and external validation sets
  993. is directly comparable.
  994. \end_layout
  995. \begin_layout Standard
  996. \begin_inset Flex TODO Note (inline)
  997. status collapsed
  998. \begin_layout Plain Layout
  999. Summarize the get.best.threshold algorithm for PAM threshold selection
  1000. \end_layout
  1001. \end_inset
  1002. \end_layout
  1003. \begin_layout Standard
  1004. Six different normalization strategies were evaluated.
  1005. First, 2 well-known non-single-channel normalization methods were considered:
  1006. RMA and dChip
  1007. \begin_inset CommandInset citation
  1008. LatexCommand cite
  1009. key "Li2001,Irizarry2003a"
  1010. literal "false"
  1011. \end_inset
  1012. .
  1013. Since RMA produces expression values on a log2 scale and dChip does not,
  1014. the values from dChip were log2 transformed after normalization.
  1015. Next, RMA and dChip followed by Global Rank-invariant Set Normalization
  1016. (GRSN) were tested
  1017. \begin_inset CommandInset citation
  1018. LatexCommand cite
  1019. key "Pelz2008"
  1020. literal "false"
  1021. \end_inset
  1022. .
  1023. Post-processing with GRSN does not turn RMA or dChip into single-channel
  1024. methods, but it may help mitigate batch effects and is therefore useful
  1025. as a benchmark.
  1026. Lastly, the two single-channel normalization methods, fRMA and SCAN, were
  1027. tested
  1028. \begin_inset CommandInset citation
  1029. LatexCommand cite
  1030. key "McCall2010,Piccolo2012"
  1031. literal "false"
  1032. \end_inset
  1033. .
  1034. When evaluting internal validation performance, only the 157 internal samples
  1035. were normalized; when evaluating external validation performance, all 157
  1036. internal samples and 75 external samples were normalized together.
  1037. \end_layout
  1038. \begin_layout Standard
  1039. For demonstrating the problem with separate normalization of training and
  1040. validation data, one additional normalization was performed: the internal
  1041. and external sets were each normalized separately using RMA, and the normalized
  1042. data for each set were combined into a single set with no further attempts
  1043. at normalizing between the two sets.
  1044. The represents approximately how RMA would have to be used in a clinical
  1045. setting, where the samples to be classified are not available at the time
  1046. the classifier is trained.
  1047. \end_layout
  1048. \begin_layout Subsection
  1049. Generating custom fRMA vectors for hthgu133pluspm array platform
  1050. \end_layout
  1051. \begin_layout Standard
  1052. In order to enable fRMA normalization for the hthgu133pluspm array platform,
  1053. custom fRMA normalization vectors were trained using the frmaTools package
  1054. \begin_inset CommandInset citation
  1055. LatexCommand cite
  1056. key "McCall2011"
  1057. literal "false"
  1058. \end_inset
  1059. .
  1060. Separate vectors were created for two types of samples: kidney graft biopsy
  1061. samples and blood samples from graft recipients.
  1062. For training, a 341 kidney biopsy samples from 2 data sets and 965 blood
  1063. samples from 5 data sets were used as the reference set.
  1064. Arrays were groups into batches based on unique combinations of sample
  1065. type (blood or biopsy), diagnosis (TX, AR, etc.), data set, and scan date.
  1066. Thus, each batch represents arrays of the same kind that were run together
  1067. on the same day.
  1068. For estimating the probe inverse variance weights, frmaTools requires equal-siz
  1069. ed batches, which means a batch size must be chosen, and then batches smaller
  1070. than that size must be ignored, while batches larger than the chosen size
  1071. must be downsampled.
  1072. This downsampling is performed randomly, so the sampling process is repeated
  1073. 5 times and the resulting normalizations are compared to each other.
  1074. \end_layout
  1075. \begin_layout Standard
  1076. To evaluate the consistency of the generated normalization vectors, the
  1077. 5 fRMA vector sets generated from 5 random batch samplings were each used
  1078. to normalize the same 20 randomly selected samples from each tissue.
  1079. Then the normalized expression values for each probe on each array were
  1080. compared across all normalizations.
  1081. Each fRMA normalization was also compared against the normalized expression
  1082. values obtained by normalizing the same 20 samples with ordinary RMA.
  1083. \end_layout
  1084. \begin_layout Subsection
  1085. Modeling methylation array M-value heteroskedasticy with modified voom implement
  1086. ation
  1087. \end_layout
  1088. \begin_layout Itemize
  1089. Methylation arrays for differential methylation in rejection vs.
  1090. healthy transplant
  1091. \end_layout
  1092. \begin_layout Itemize
  1093. Adapt voom method originally designed for RNA-seq to model mean-variance
  1094. dependence
  1095. \end_layout
  1096. \begin_layout Itemize
  1097. Use sample precision weighting, duplicateCorrelation, and sva to adjust
  1098. for other confounding factors
  1099. \end_layout
  1100. \begin_layout Section
  1101. Results
  1102. \end_layout
  1103. \begin_layout Standard
  1104. \begin_inset Flex TODO Note (inline)
  1105. status open
  1106. \begin_layout Plain Layout
  1107. Improve subsection titles in this section
  1108. \end_layout
  1109. \end_inset
  1110. \end_layout
  1111. \begin_layout Subsection
  1112. fRMA eliminates unwanted dependence of classifier training on normalization
  1113. strategy caused by RMA
  1114. \end_layout
  1115. \begin_layout Subsubsection
  1116. Separate normalization with RMA introduces unwanted biases in classification
  1117. \end_layout
  1118. \begin_layout Standard
  1119. \begin_inset Float figure
  1120. wide false
  1121. sideways false
  1122. status collapsed
  1123. \begin_layout Plain Layout
  1124. \begin_inset Graphics
  1125. filename graphics/PAM/predplot.pdf
  1126. \end_inset
  1127. \end_layout
  1128. \begin_layout Plain Layout
  1129. \begin_inset Caption Standard
  1130. \begin_layout Plain Layout
  1131. \begin_inset CommandInset label
  1132. LatexCommand label
  1133. name "fig:Classifier-probabilities-RMA"
  1134. \end_inset
  1135. \series bold
  1136. Classifier probabilities on validation samples when normalized with RMA
  1137. together vs.
  1138. separately.
  1139. \end_layout
  1140. \end_inset
  1141. \end_layout
  1142. \end_inset
  1143. \end_layout
  1144. \begin_layout Standard
  1145. To demonstrate the problem with non-single-channel methods, we considered
  1146. the problem of training a classifier to distinguish TX from AR using the
  1147. samples from the internal set as training data, evaluating performance
  1148. on the external set.
  1149. First, training and evaluation were performed after normalizing all array
  1150. samples together as a single set using RMA, and second, the internal samples
  1151. were normalized separately from the external samples and the training and
  1152. evaluation were repeated.
  1153. For each sample in the validation set, the classifier probabilities from
  1154. both classifiers were plotted against each other (Fig.
  1155. \begin_inset CommandInset ref
  1156. LatexCommand ref
  1157. reference "fig:Classifier-probabilities-RMA"
  1158. plural "false"
  1159. caps "false"
  1160. noprefix "false"
  1161. \end_inset
  1162. ).
  1163. As expected, separate normalization biases the classifier probabilities,
  1164. resulting in several misclassifications.
  1165. In this case, the bias from separate normalization causes the classifier
  1166. to assign a lower probability of AR to every sample.
  1167. \end_layout
  1168. \begin_layout Subsubsection
  1169. fRMA and SCAN achieve maintain classification performance while eliminating
  1170. dependence on normalization strategy
  1171. \end_layout
  1172. \begin_layout Standard
  1173. \begin_inset Float figure
  1174. wide false
  1175. sideways false
  1176. status collapsed
  1177. \begin_layout Plain Layout
  1178. \begin_inset Graphics
  1179. filename graphics/PAM/ROC-TXvsAR-internal.pdf
  1180. width 100col%
  1181. groupId colwidth
  1182. \end_inset
  1183. \end_layout
  1184. \begin_layout Plain Layout
  1185. \begin_inset Caption Standard
  1186. \begin_layout Plain Layout
  1187. \begin_inset CommandInset label
  1188. LatexCommand label
  1189. name "fig:ROC-PAM-int"
  1190. \end_inset
  1191. ROC curves for PAM on internal validation data using different normalization
  1192. strategies
  1193. \end_layout
  1194. \end_inset
  1195. \end_layout
  1196. \end_inset
  1197. \end_layout
  1198. \begin_layout Standard
  1199. \begin_inset Float table
  1200. wide false
  1201. sideways false
  1202. status collapsed
  1203. \begin_layout Plain Layout
  1204. \begin_inset Tabular
  1205. <lyxtabular version="3" rows="7" columns="4">
  1206. <features tabularvalignment="middle">
  1207. <column alignment="center" valignment="top">
  1208. <column alignment="center" valignment="top">
  1209. <column alignment="center" valignment="top">
  1210. <column alignment="center" valignment="top">
  1211. <row>
  1212. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  1213. \begin_inset Text
  1214. \begin_layout Plain Layout
  1215. \family roman
  1216. \series medium
  1217. \shape up
  1218. \size normal
  1219. \emph off
  1220. \bar no
  1221. \strikeout off
  1222. \xout off
  1223. \uuline off
  1224. \uwave off
  1225. \noun off
  1226. \color none
  1227. Normalization
  1228. \end_layout
  1229. \end_inset
  1230. </cell>
  1231. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  1232. \begin_inset Text
  1233. \begin_layout Plain Layout
  1234. Single-channel
  1235. \end_layout
  1236. \end_inset
  1237. </cell>
  1238. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  1239. \begin_inset Text
  1240. \begin_layout Plain Layout
  1241. \family roman
  1242. \series medium
  1243. \shape up
  1244. \size normal
  1245. \emph off
  1246. \bar no
  1247. \strikeout off
  1248. \xout off
  1249. \uuline off
  1250. \uwave off
  1251. \noun off
  1252. \color none
  1253. Internal Validation AUC
  1254. \end_layout
  1255. \end_inset
  1256. </cell>
  1257. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  1258. \begin_inset Text
  1259. \begin_layout Plain Layout
  1260. External Validation AUC
  1261. \end_layout
  1262. \end_inset
  1263. </cell>
  1264. </row>
  1265. <row>
  1266. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1267. \begin_inset Text
  1268. \begin_layout Plain Layout
  1269. \family roman
  1270. \series medium
  1271. \shape up
  1272. \size normal
  1273. \emph off
  1274. \bar no
  1275. \strikeout off
  1276. \xout off
  1277. \uuline off
  1278. \uwave off
  1279. \noun off
  1280. \color none
  1281. RMA
  1282. \end_layout
  1283. \end_inset
  1284. </cell>
  1285. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1286. \begin_inset Text
  1287. \begin_layout Plain Layout
  1288. No
  1289. \end_layout
  1290. \end_inset
  1291. </cell>
  1292. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1293. \begin_inset Text
  1294. \begin_layout Plain Layout
  1295. \family roman
  1296. \series medium
  1297. \shape up
  1298. \size normal
  1299. \emph off
  1300. \bar no
  1301. \strikeout off
  1302. \xout off
  1303. \uuline off
  1304. \uwave off
  1305. \noun off
  1306. \color none
  1307. 0.852
  1308. \end_layout
  1309. \end_inset
  1310. </cell>
  1311. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  1312. \begin_inset Text
  1313. \begin_layout Plain Layout
  1314. \family roman
  1315. \series medium
  1316. \shape up
  1317. \size normal
  1318. \emph off
  1319. \bar no
  1320. \strikeout off
  1321. \xout off
  1322. \uuline off
  1323. \uwave off
  1324. \noun off
  1325. \color none
  1326. 0.713
  1327. \end_layout
  1328. \end_inset
  1329. </cell>
  1330. </row>
  1331. <row>
  1332. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1333. \begin_inset Text
  1334. \begin_layout Plain Layout
  1335. \family roman
  1336. \series medium
  1337. \shape up
  1338. \size normal
  1339. \emph off
  1340. \bar no
  1341. \strikeout off
  1342. \xout off
  1343. \uuline off
  1344. \uwave off
  1345. \noun off
  1346. \color none
  1347. dChip
  1348. \end_layout
  1349. \end_inset
  1350. </cell>
  1351. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1352. \begin_inset Text
  1353. \begin_layout Plain Layout
  1354. No
  1355. \end_layout
  1356. \end_inset
  1357. </cell>
  1358. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1359. \begin_inset Text
  1360. \begin_layout Plain Layout
  1361. \family roman
  1362. \series medium
  1363. \shape up
  1364. \size normal
  1365. \emph off
  1366. \bar no
  1367. \strikeout off
  1368. \xout off
  1369. \uuline off
  1370. \uwave off
  1371. \noun off
  1372. \color none
  1373. 0.891
  1374. \end_layout
  1375. \end_inset
  1376. </cell>
  1377. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  1378. \begin_inset Text
  1379. \begin_layout Plain Layout
  1380. \family roman
  1381. \series medium
  1382. \shape up
  1383. \size normal
  1384. \emph off
  1385. \bar no
  1386. \strikeout off
  1387. \xout off
  1388. \uuline off
  1389. \uwave off
  1390. \noun off
  1391. \color none
  1392. 0.657
  1393. \end_layout
  1394. \end_inset
  1395. </cell>
  1396. </row>
  1397. <row>
  1398. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1399. \begin_inset Text
  1400. \begin_layout Plain Layout
  1401. \family roman
  1402. \series medium
  1403. \shape up
  1404. \size normal
  1405. \emph off
  1406. \bar no
  1407. \strikeout off
  1408. \xout off
  1409. \uuline off
  1410. \uwave off
  1411. \noun off
  1412. \color none
  1413. RMA + GRSN
  1414. \end_layout
  1415. \end_inset
  1416. </cell>
  1417. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1418. \begin_inset Text
  1419. \begin_layout Plain Layout
  1420. No
  1421. \end_layout
  1422. \end_inset
  1423. </cell>
  1424. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1425. \begin_inset Text
  1426. \begin_layout Plain Layout
  1427. \family roman
  1428. \series medium
  1429. \shape up
  1430. \size normal
  1431. \emph off
  1432. \bar no
  1433. \strikeout off
  1434. \xout off
  1435. \uuline off
  1436. \uwave off
  1437. \noun off
  1438. \color none
  1439. 0.816
  1440. \end_layout
  1441. \end_inset
  1442. </cell>
  1443. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  1444. \begin_inset Text
  1445. \begin_layout Plain Layout
  1446. \family roman
  1447. \series medium
  1448. \shape up
  1449. \size normal
  1450. \emph off
  1451. \bar no
  1452. \strikeout off
  1453. \xout off
  1454. \uuline off
  1455. \uwave off
  1456. \noun off
  1457. \color none
  1458. 0.750
  1459. \end_layout
  1460. \end_inset
  1461. </cell>
  1462. </row>
  1463. <row>
  1464. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1465. \begin_inset Text
  1466. \begin_layout Plain Layout
  1467. \family roman
  1468. \series medium
  1469. \shape up
  1470. \size normal
  1471. \emph off
  1472. \bar no
  1473. \strikeout off
  1474. \xout off
  1475. \uuline off
  1476. \uwave off
  1477. \noun off
  1478. \color none
  1479. dChip + GRSN
  1480. \end_layout
  1481. \end_inset
  1482. </cell>
  1483. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1484. \begin_inset Text
  1485. \begin_layout Plain Layout
  1486. No
  1487. \end_layout
  1488. \end_inset
  1489. </cell>
  1490. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1491. \begin_inset Text
  1492. \begin_layout Plain Layout
  1493. \family roman
  1494. \series medium
  1495. \shape up
  1496. \size normal
  1497. \emph off
  1498. \bar no
  1499. \strikeout off
  1500. \xout off
  1501. \uuline off
  1502. \uwave off
  1503. \noun off
  1504. \color none
  1505. 0.875
  1506. \end_layout
  1507. \end_inset
  1508. </cell>
  1509. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  1510. \begin_inset Text
  1511. \begin_layout Plain Layout
  1512. \family roman
  1513. \series medium
  1514. \shape up
  1515. \size normal
  1516. \emph off
  1517. \bar no
  1518. \strikeout off
  1519. \xout off
  1520. \uuline off
  1521. \uwave off
  1522. \noun off
  1523. \color none
  1524. 0.642
  1525. \end_layout
  1526. \end_inset
  1527. </cell>
  1528. </row>
  1529. <row>
  1530. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1531. \begin_inset Text
  1532. \begin_layout Plain Layout
  1533. \family roman
  1534. \series medium
  1535. \shape up
  1536. \size normal
  1537. \emph off
  1538. \bar no
  1539. \strikeout off
  1540. \xout off
  1541. \uuline off
  1542. \uwave off
  1543. \noun off
  1544. \color none
  1545. fRMA
  1546. \end_layout
  1547. \end_inset
  1548. </cell>
  1549. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1550. \begin_inset Text
  1551. \begin_layout Plain Layout
  1552. Yes
  1553. \end_layout
  1554. \end_inset
  1555. </cell>
  1556. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  1557. \begin_inset Text
  1558. \begin_layout Plain Layout
  1559. \family roman
  1560. \series medium
  1561. \shape up
  1562. \size normal
  1563. \emph off
  1564. \bar no
  1565. \strikeout off
  1566. \xout off
  1567. \uuline off
  1568. \uwave off
  1569. \noun off
  1570. \color none
  1571. 0.863
  1572. \end_layout
  1573. \end_inset
  1574. </cell>
  1575. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  1576. \begin_inset Text
  1577. \begin_layout Plain Layout
  1578. \family roman
  1579. \series medium
  1580. \shape up
  1581. \size normal
  1582. \emph off
  1583. \bar no
  1584. \strikeout off
  1585. \xout off
  1586. \uuline off
  1587. \uwave off
  1588. \noun off
  1589. \color none
  1590. 0.718
  1591. \end_layout
  1592. \end_inset
  1593. </cell>
  1594. </row>
  1595. <row>
  1596. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  1597. \begin_inset Text
  1598. \begin_layout Plain Layout
  1599. \family roman
  1600. \series medium
  1601. \shape up
  1602. \size normal
  1603. \emph off
  1604. \bar no
  1605. \strikeout off
  1606. \xout off
  1607. \uuline off
  1608. \uwave off
  1609. \noun off
  1610. \color none
  1611. SCAN
  1612. \end_layout
  1613. \end_inset
  1614. </cell>
  1615. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  1616. \begin_inset Text
  1617. \begin_layout Plain Layout
  1618. Yes
  1619. \end_layout
  1620. \end_inset
  1621. </cell>
  1622. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  1623. \begin_inset Text
  1624. \begin_layout Plain Layout
  1625. \family roman
  1626. \series medium
  1627. \shape up
  1628. \size normal
  1629. \emph off
  1630. \bar no
  1631. \strikeout off
  1632. \xout off
  1633. \uuline off
  1634. \uwave off
  1635. \noun off
  1636. \color none
  1637. 0.853
  1638. \end_layout
  1639. \end_inset
  1640. </cell>
  1641. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  1642. \begin_inset Text
  1643. \begin_layout Plain Layout
  1644. \family roman
  1645. \series medium
  1646. \shape up
  1647. \size normal
  1648. \emph off
  1649. \bar no
  1650. \strikeout off
  1651. \xout off
  1652. \uuline off
  1653. \uwave off
  1654. \noun off
  1655. \color none
  1656. 0.689
  1657. \end_layout
  1658. \end_inset
  1659. </cell>
  1660. </row>
  1661. </lyxtabular>
  1662. \end_inset
  1663. \end_layout
  1664. \begin_layout Plain Layout
  1665. \begin_inset Caption Standard
  1666. \begin_layout Plain Layout
  1667. \begin_inset CommandInset label
  1668. LatexCommand label
  1669. name "tab:AUC-PAM"
  1670. \end_inset
  1671. \series bold
  1672. AUC values for internal and external validation with 6 different normalization
  1673. strategies.
  1674. \series default
  1675. Only fRMA and SCAN are single-channel normalizations.
  1676. The other 4 normalizations are for comparison.
  1677. \end_layout
  1678. \end_inset
  1679. \end_layout
  1680. \end_inset
  1681. \end_layout
  1682. \begin_layout Standard
  1683. For internal validation, the 6 methods' AUC values ranged from 0.816 to 0.891,
  1684. as shown in Table
  1685. \begin_inset CommandInset ref
  1686. LatexCommand ref
  1687. reference "tab:AUC-PAM"
  1688. plural "false"
  1689. caps "false"
  1690. noprefix "false"
  1691. \end_inset
  1692. .
  1693. Among the non-single-channel normalizations, dChip outperformed RMA, while
  1694. GRSN reduced the AUC values for both dChip and RMA.
  1695. Both single-channel methods, fRMA and SCAN, slightly outperformed RMA,
  1696. with fRMA ahead of SCAN.
  1697. However, the difference between RMA and fRMA is still quite small.
  1698. Figure
  1699. \begin_inset CommandInset ref
  1700. LatexCommand ref
  1701. reference "fig:ROC-PAM-int"
  1702. plural "false"
  1703. caps "false"
  1704. noprefix "false"
  1705. \end_inset
  1706. shows that the ROC curves for RMA, dChip, and fRMA look very similar and
  1707. relatively smooth, while both GRSN curves and the curve for SCAN have a
  1708. more jagged appearance.
  1709. \end_layout
  1710. \begin_layout Standard
  1711. \begin_inset Float figure
  1712. wide false
  1713. sideways false
  1714. status collapsed
  1715. \begin_layout Plain Layout
  1716. \begin_inset Graphics
  1717. filename graphics/PAM/ROC-TXvsAR-external.pdf
  1718. width 100col%
  1719. groupId colwidth
  1720. \end_inset
  1721. \end_layout
  1722. \begin_layout Plain Layout
  1723. \begin_inset Caption Standard
  1724. \begin_layout Plain Layout
  1725. \begin_inset CommandInset label
  1726. LatexCommand label
  1727. name "fig:ROC-PAM-ext"
  1728. \end_inset
  1729. ROC curve for PAM on external validation data using different normalization
  1730. strategies
  1731. \end_layout
  1732. \end_inset
  1733. \end_layout
  1734. \end_inset
  1735. \end_layout
  1736. \begin_layout Standard
  1737. For external validation, as expected, all the AUC values are lower than
  1738. the internal validations, ranging from 0.642 to 0.750 (Table
  1739. \begin_inset CommandInset ref
  1740. LatexCommand ref
  1741. reference "tab:AUC-PAM"
  1742. plural "false"
  1743. caps "false"
  1744. noprefix "false"
  1745. \end_inset
  1746. ).
  1747. With or without GRSN, RMA shows its dominance over dChip in this more challengi
  1748. ng test.
  1749. Unlike in the internal validation, GRSN actually improves the classifier
  1750. performance for RMA, although it does not for dChip.
  1751. Once again, both single-channel methods perform about on par with RMA,
  1752. with fRMA performing slightly better and SCAN performing a bit worse.
  1753. Figure
  1754. \begin_inset CommandInset ref
  1755. LatexCommand ref
  1756. reference "fig:ROC-PAM-ext"
  1757. plural "false"
  1758. caps "false"
  1759. noprefix "false"
  1760. \end_inset
  1761. shows the ROC curves for the external validation test.
  1762. As expected, none of them are as clean-looking as the internal validation
  1763. ROC curves.
  1764. The curves for RMA, RMA+GRSN, and fRMA all look similar, while the other
  1765. curves look more divergent.
  1766. \end_layout
  1767. \begin_layout Subsection
  1768. fRMA with custom-generated vectors enables normalization on hthgu133pluspm
  1769. \end_layout
  1770. \begin_layout Standard
  1771. \begin_inset Float figure
  1772. wide false
  1773. sideways false
  1774. status open
  1775. \begin_layout Plain Layout
  1776. \begin_inset Graphics
  1777. filename graphics/frma-pax-bx/batchsize_batches.pdf
  1778. \end_inset
  1779. \end_layout
  1780. \begin_layout Plain Layout
  1781. \begin_inset Caption Standard
  1782. \begin_layout Plain Layout
  1783. \begin_inset CommandInset label
  1784. LatexCommand label
  1785. name "fig:batch-size-batches"
  1786. \end_inset
  1787. \series bold
  1788. Effect of batch size selection on number of batches included in fRMA probe
  1789. weight learning.
  1790. \series default
  1791. For batch sizes ranging from 3 to 15, the number of batches with at least
  1792. that many samples was plotted for biopsy (BX) and blood (PAX) samples.
  1793. The selected batch size, 5, is marked with a dotted vertical line.
  1794. \end_layout
  1795. \end_inset
  1796. \end_layout
  1797. \end_inset
  1798. \end_layout
  1799. \begin_layout Standard
  1800. \begin_inset Float figure
  1801. wide false
  1802. sideways false
  1803. status open
  1804. \begin_layout Plain Layout
  1805. \begin_inset Graphics
  1806. filename graphics/frma-pax-bx/batchsize_samples.pdf
  1807. \end_inset
  1808. \end_layout
  1809. \begin_layout Plain Layout
  1810. \begin_inset Caption Standard
  1811. \begin_layout Plain Layout
  1812. \begin_inset CommandInset label
  1813. LatexCommand label
  1814. name "fig:batch-size-samples"
  1815. \end_inset
  1816. \series bold
  1817. Effect of batch size selection on number of samples included in fRMA probe
  1818. weight learning.
  1819. \series default
  1820. For batch sizes ranging from 3 to 15, the number of samples included in
  1821. probe weight training was plotted for biopsy (BX) and blood (PAX) samples.
  1822. The selected batch size, 5, is marked with a dotted vertical line.
  1823. \end_layout
  1824. \end_inset
  1825. \end_layout
  1826. \end_inset
  1827. \end_layout
  1828. \begin_layout Standard
  1829. In order to enable use of fRMA to normalize hthgu133pluspm, a custom set
  1830. of fRMA vectors was created.
  1831. First, an appropriate batch size was chosen by looking at the number of
  1832. batches and number of samples included as a function of batch size (Figures
  1833. \begin_inset CommandInset ref
  1834. LatexCommand ref
  1835. reference "fig:batch-size-batches"
  1836. plural "false"
  1837. caps "false"
  1838. noprefix "false"
  1839. \end_inset
  1840. and
  1841. \begin_inset CommandInset ref
  1842. LatexCommand ref
  1843. reference "fig:batch-size-samples"
  1844. plural "false"
  1845. caps "false"
  1846. noprefix "false"
  1847. \end_inset
  1848. , respectively).
  1849. For a given batch size, all batches with fewer samples that the chosen
  1850. size must be ignored during training, while larger batches must be randomly
  1851. downsampled to the chosen size.
  1852. Hence, the number of samples included for a given batch size equals the
  1853. batch size times the number of batches with at least that many samples.
  1854. From Figure
  1855. \begin_inset CommandInset ref
  1856. LatexCommand ref
  1857. reference "fig:batch-size-samples"
  1858. plural "false"
  1859. caps "false"
  1860. noprefix "false"
  1861. \end_inset
  1862. , it is apparent that that a batch size of 8 maximizes the number of samples
  1863. included in training.
  1864. Increasing the batch size beyond this causes too many smaller batches to
  1865. be excluded, reducing the total number of samples for both tissue types.
  1866. However, a batch size of 8 is not necessarily optimal.
  1867. The article introducing frmaTools concluded that it was highly advantageous
  1868. to use a smaller batch size in order to include more batches, even at the
  1869. expense of including fewer total samples in training
  1870. \begin_inset CommandInset citation
  1871. LatexCommand cite
  1872. key "McCall2011"
  1873. literal "false"
  1874. \end_inset
  1875. .
  1876. To strike an appropriate balance between more batches and more samples,
  1877. a batch size of 5 was chosen.
  1878. For both blood and biopsy samples, this increased the number of batches
  1879. included by 10, with only a modest reduction in the number of samples compared
  1880. to a batch size of 8.
  1881. With a batch size of 5, 26 batches of biopsy samples and 46 batches of
  1882. blood samples were available.
  1883. \end_layout
  1884. \begin_layout Standard
  1885. \begin_inset Float figure
  1886. wide false
  1887. sideways false
  1888. status collapsed
  1889. \begin_layout Plain Layout
  1890. \begin_inset Graphics
  1891. filename graphics/frma-pax-bx/M-BX-violin.pdf
  1892. lyxscale 30
  1893. groupId m-violin
  1894. \end_inset
  1895. \end_layout
  1896. \begin_layout Plain Layout
  1897. \begin_inset Caption Standard
  1898. \begin_layout Plain Layout
  1899. \begin_inset CommandInset label
  1900. LatexCommand label
  1901. name "fig:m-bx-violin"
  1902. \end_inset
  1903. \series bold
  1904. Violin plot of log ratios between normalizations for 20 biopsy samples.
  1905. \series default
  1906. Each of 20 randomly selected biopsy samples was normalized with RMA and
  1907. with 5 different sets of fRMA vectors.
  1908. This shows the distribution of log ratios between normalized expression
  1909. values, aggregated across all 20 arrays.
  1910. \end_layout
  1911. \end_inset
  1912. \end_layout
  1913. \end_inset
  1914. \end_layout
  1915. \begin_layout Standard
  1916. Since fRMA training requires equal-size batches, larger batches are downsampled
  1917. randomly.
  1918. This introduces a nondeterministic step in the generation of normalization
  1919. vectors.
  1920. To show that this randomness does not substantially change the outcome,
  1921. the random downsampling and subsequent vector learning was repeated 5 times,
  1922. with a different random seed each time.
  1923. 20 samples were selected at random as a test set and normalized with each
  1924. of the 5 sets of fRMA normalization vectors as well as ordinary RMA, and
  1925. the normalized expression values were compared across normalizations.
  1926. Figure
  1927. \begin_inset CommandInset ref
  1928. LatexCommand ref
  1929. reference "fig:m-bx-violin"
  1930. plural "false"
  1931. caps "false"
  1932. noprefix "false"
  1933. \end_inset
  1934. shows a summary of these comparisons for biopsy samples.
  1935. Comparing RMA to each of the 5 fRMA normalizations, the distribution of
  1936. log ratios is somewhat wide, indicating that the normalizations disagree
  1937. on the expression values of a fair number of probe sets.
  1938. In contrast, comparisons of fRMA against fRMA, the vast mojority of probe
  1939. sets have very small log ratios, indicating a very high agreement between
  1940. the normalized values generated by the two normalizations.
  1941. This shows that the fRMA normalization's behavior is not very sensitive
  1942. to the random downsampling of larger batches during training.
  1943. \end_layout
  1944. \begin_layout Standard
  1945. \begin_inset Float figure
  1946. wide false
  1947. sideways false
  1948. status collapsed
  1949. \begin_layout Plain Layout
  1950. \begin_inset Graphics
  1951. filename graphics/frma-pax-bx/MA-BX-RMA.fRMA.pdf
  1952. lyxscale 50
  1953. groupId ma-frma
  1954. \end_inset
  1955. \end_layout
  1956. \begin_layout Plain Layout
  1957. \begin_inset Caption Standard
  1958. \begin_layout Plain Layout
  1959. \begin_inset CommandInset label
  1960. LatexCommand label
  1961. name "fig:ma-bx-rma-frma"
  1962. \end_inset
  1963. \series bold
  1964. Representative MA plot comparing RMA against fRMA for 20 biopsy samples.
  1965. \series default
  1966. Averages and log ratios were computed for every probe in each of 20 biopsy
  1967. samples between RMA normalization and fRMA.
  1968. Density of points is represented by darkness of shading, and individual
  1969. outlier points are plotted.
  1970. \end_layout
  1971. \end_inset
  1972. \end_layout
  1973. \end_inset
  1974. \end_layout
  1975. \begin_layout Standard
  1976. \begin_inset Float figure
  1977. wide false
  1978. sideways false
  1979. status collapsed
  1980. \begin_layout Plain Layout
  1981. \begin_inset Graphics
  1982. filename graphics/frma-pax-bx/MA-BX-fRMA.fRMA.pdf
  1983. lyxscale 50
  1984. groupId ma-frma
  1985. \end_inset
  1986. \end_layout
  1987. \begin_layout Plain Layout
  1988. \begin_inset Caption Standard
  1989. \begin_layout Plain Layout
  1990. \begin_inset CommandInset label
  1991. LatexCommand label
  1992. name "fig:ma-bx-frma-frma"
  1993. \end_inset
  1994. \series bold
  1995. Representative MA plot comparing different fRMA vectors for 20 biopsy samples.
  1996. \series default
  1997. Averages and log ratios were computed for every probe in each of 20 biopsy
  1998. samples between fRMA normalizations using vectors from two different batch
  1999. samplings.
  2000. Density of points is represented by darkness of shading, and individual
  2001. outlier points are plotted.
  2002. \end_layout
  2003. \end_inset
  2004. \end_layout
  2005. \end_inset
  2006. \end_layout
  2007. \begin_layout Standard
  2008. Figure
  2009. \begin_inset CommandInset ref
  2010. LatexCommand ref
  2011. reference "fig:ma-bx-rma-frma"
  2012. plural "false"
  2013. caps "false"
  2014. noprefix "false"
  2015. \end_inset
  2016. shows an MA plot of the RMA-normalized values against the fRMA-normalized
  2017. values for the same probe sets and arrays, corresponding to the first row
  2018. of Figure
  2019. \begin_inset CommandInset ref
  2020. LatexCommand ref
  2021. reference "fig:m-bx-violin"
  2022. plural "false"
  2023. caps "false"
  2024. noprefix "false"
  2025. \end_inset
  2026. .
  2027. This MA plot shows that not only is there a wide distribution of M-values,
  2028. but the trend of M-values is dependent on the average normalized intensity.
  2029. This is expected, since the overall trend represents the differences in
  2030. the quantile normalization step.
  2031. When running RMA, only the quantiles for these specific 20 arrays are used,
  2032. while for fRMA the quantile distribution is taking from all arrays used
  2033. in training.
  2034. Figure
  2035. \begin_inset CommandInset ref
  2036. LatexCommand ref
  2037. reference "fig:ma-bx-frma-frma"
  2038. plural "false"
  2039. caps "false"
  2040. noprefix "false"
  2041. \end_inset
  2042. shows a similar MA plot comparing 2 different fRMA normalizations, correspondin
  2043. g to the 6th row of Figure
  2044. \begin_inset CommandInset ref
  2045. LatexCommand ref
  2046. reference "fig:m-bx-violin"
  2047. plural "false"
  2048. caps "false"
  2049. noprefix "false"
  2050. \end_inset
  2051. .
  2052. The MA plot is very tightly centered around zero with no visible trend.
  2053. Figures
  2054. \begin_inset CommandInset ref
  2055. LatexCommand ref
  2056. reference "fig:m-pax-violin"
  2057. plural "false"
  2058. caps "false"
  2059. noprefix "false"
  2060. \end_inset
  2061. ,
  2062. \begin_inset CommandInset ref
  2063. LatexCommand ref
  2064. reference "fig:MA-PAX-rma-frma"
  2065. plural "false"
  2066. caps "false"
  2067. noprefix "false"
  2068. \end_inset
  2069. , and
  2070. \begin_inset CommandInset ref
  2071. LatexCommand ref
  2072. reference "fig:ma-bx-frma-frma"
  2073. plural "false"
  2074. caps "false"
  2075. noprefix "false"
  2076. \end_inset
  2077. show exactly the same information for the blood samples, once again comparing
  2078. the normalized expression values between normalizations for all probe sets
  2079. across 20 randomly selected test arrays.
  2080. Once again, there is a wider distribution of log ratios between RMA-normalized
  2081. values and fRMA-normalized, and a much tighter distribution when comparing
  2082. different fRMA normalizations to each other, indicating that the fRMA training
  2083. process is robust to random batch downsampling for the blood samples as
  2084. well.
  2085. \end_layout
  2086. \begin_layout Standard
  2087. \begin_inset Float figure
  2088. wide false
  2089. sideways false
  2090. status collapsed
  2091. \begin_layout Plain Layout
  2092. \begin_inset Graphics
  2093. filename graphics/frma-pax-bx/M-PAX-violin.pdf
  2094. lyxscale 30
  2095. groupId m-violin
  2096. \end_inset
  2097. \end_layout
  2098. \begin_layout Plain Layout
  2099. \begin_inset Caption Standard
  2100. \begin_layout Plain Layout
  2101. \begin_inset CommandInset label
  2102. LatexCommand label
  2103. name "fig:m-pax-violin"
  2104. \end_inset
  2105. \series bold
  2106. Violin plot of log ratios between normalizations for 20 blood samples.
  2107. \series default
  2108. Each of 20 randomly selected blood samples was normalized with RMA and with
  2109. 5 different sets of fRMA vectors.
  2110. This shows the distribution of log ratios between normalized expression
  2111. values, aggregated across all 20 arrays.
  2112. \end_layout
  2113. \end_inset
  2114. \end_layout
  2115. \end_inset
  2116. \end_layout
  2117. \begin_layout Standard
  2118. \begin_inset Float figure
  2119. wide false
  2120. sideways false
  2121. status collapsed
  2122. \begin_layout Plain Layout
  2123. \begin_inset Graphics
  2124. filename graphics/frma-pax-bx/MA-PAX-RMA.fRMA.pdf
  2125. lyxscale 50
  2126. groupId ma-frma
  2127. \end_inset
  2128. \end_layout
  2129. \begin_layout Plain Layout
  2130. \begin_inset Caption Standard
  2131. \begin_layout Plain Layout
  2132. \begin_inset CommandInset label
  2133. LatexCommand label
  2134. name "fig:MA-PAX-rma-frma"
  2135. \end_inset
  2136. \series bold
  2137. Representative MA plot comparing RMA against fRMA for 20 blood samples.
  2138. \series default
  2139. Averages and log ratios were computed for every probe in each of 20 blood
  2140. samples between RMA normalization and fRMA.
  2141. Density of points is represented by darkness of shading, and individual
  2142. outlier points are plotted.
  2143. \end_layout
  2144. \end_inset
  2145. \end_layout
  2146. \begin_layout Plain Layout
  2147. \end_layout
  2148. \end_inset
  2149. \end_layout
  2150. \begin_layout Standard
  2151. \begin_inset Float figure
  2152. wide false
  2153. sideways false
  2154. status collapsed
  2155. \begin_layout Plain Layout
  2156. \begin_inset Graphics
  2157. filename graphics/frma-pax-bx/MA-PAX-fRMA.fRMA.pdf
  2158. lyxscale 50
  2159. groupId ma-frma
  2160. \end_inset
  2161. \end_layout
  2162. \begin_layout Plain Layout
  2163. \begin_inset Caption Standard
  2164. \begin_layout Plain Layout
  2165. \begin_inset CommandInset label
  2166. LatexCommand label
  2167. name "fig:MA-PAX-frma-frma"
  2168. \end_inset
  2169. \series bold
  2170. Representative MA plot comparing different fRMA vectors for 20 blood samples.
  2171. \series default
  2172. Averages and log ratios were computed for every probe in each of 20 blood
  2173. samples between fRMA normalizations using vectors from two different batch
  2174. samplings.
  2175. Density of points is represented by darkness of shading, and individual
  2176. outlier points are plotted.
  2177. \end_layout
  2178. \end_inset
  2179. \end_layout
  2180. \end_inset
  2181. \end_layout
  2182. \begin_layout Subsection
  2183. Adapting voom to methylation array data improves model fit
  2184. \end_layout
  2185. \begin_layout Itemize
  2186. voom, precision weights, and sva improved model fit
  2187. \end_layout
  2188. \begin_deeper
  2189. \begin_layout Itemize
  2190. Also increased sensitivity for detecting differential methylation
  2191. \end_layout
  2192. \end_deeper
  2193. \begin_layout Itemize
  2194. Figure showing (a) heteroskedasticy without voom, (b) voom-modeled mean-variance
  2195. trend, and (c) homoskedastic mean-variance trend after running voom
  2196. \end_layout
  2197. \begin_layout Itemize
  2198. Figure showing sample weights and their relations to
  2199. \end_layout
  2200. \begin_layout Itemize
  2201. Figure showing MDS plot with and without SVA correction
  2202. \end_layout
  2203. \begin_layout Itemize
  2204. Figure and/or table showing improved p-value historgrams/number of significant
  2205. genes (might need to get this from Padma)
  2206. \end_layout
  2207. \begin_layout Section
  2208. Discussion
  2209. \end_layout
  2210. \begin_layout Subsection
  2211. fRMA achieves clinically applicable normalization without sacrificing classifica
  2212. tion performance
  2213. \end_layout
  2214. \begin_layout Standard
  2215. As shown in Figure
  2216. \begin_inset CommandInset ref
  2217. LatexCommand ref
  2218. reference "fig:Classifier-probabilities-RMA"
  2219. plural "false"
  2220. caps "false"
  2221. noprefix "false"
  2222. \end_inset
  2223. , improper normalization, particularly separate normalization of training
  2224. and test samples, leads to unwanted biases in classification.
  2225. In a controlled experimental context, it is always possible to correct
  2226. this issue by normalizing all experimental samples together.
  2227. However, because it is not feasible to normalize all samples together in
  2228. a clinical context, a single-channel normalization is required is required.
  2229. \end_layout
  2230. \begin_layout Standard
  2231. The major concern in using a single-channel normalization is that non-single-cha
  2232. nnel methods can share information between arrays to improve the normalization,
  2233. and single-channel methods risk sacrificing the gains in normalization
  2234. accuracy that come from this information sharing.
  2235. In the case of RMA, this information sharing is accomplished through quantile
  2236. normalization and median polish steps.
  2237. The need for information sharing in quantile normalization can easily be
  2238. removed by learning a fixed set of quantiles from external data and normalizing
  2239. each array to these fixed quantiles, instead of the quantiles of the data
  2240. itself.
  2241. As long as the fixed quantiles are reasonable, the result will be similar
  2242. to standard RMA.
  2243. However, there is no analogous way to eliminate cross-array information
  2244. sharing in the median polish step, so fRMA replaces this with a weighted
  2245. average of probes on each array, with the weights learned from external
  2246. data.
  2247. This step of fRMA has the greatest potential to diverge from RMA un undesirable
  2248. ways.
  2249. \end_layout
  2250. \begin_layout Standard
  2251. However, when run on real data, fRMA performed at least as well as RMA in
  2252. both the internal validation and external validation tests.
  2253. This shows that fRMA can be used to normalize individual clinical samples
  2254. in a class prediction context without sacrificing the classifier performance
  2255. that would be obtained by using the more well-established RMA for normalization.
  2256. The other single-channel normalization method considered, SCAN, showed
  2257. some loss of AUC in the external validation test.
  2258. Based on these results, fRMA is the preferred normalization for clinical
  2259. samples in a class prediction context.
  2260. \end_layout
  2261. \begin_layout Subsection
  2262. Robust fRMA vectors can be generated for new array platforms
  2263. \end_layout
  2264. \begin_layout Standard
  2265. The published fRMA normalization vectors for the hgu133plus2 platform were
  2266. generated from a set of about 850 samples
  2267. \begin_inset Flex TODO Note (Margin)
  2268. status collapsed
  2269. \begin_layout Plain Layout
  2270. Look up the exact numbers
  2271. \end_layout
  2272. \end_inset
  2273. chosen from a wide range of tissues, which the authors determined was sufficien
  2274. t to generate a robust set of normalization vectors that could be applied
  2275. across all tissues
  2276. \begin_inset CommandInset citation
  2277. LatexCommand cite
  2278. key "McCall2010"
  2279. literal "false"
  2280. \end_inset
  2281. .
  2282. Since we only had hthgu133pluspm for 2 tissues of interest, our needs were
  2283. more modest.
  2284. Even using only 130 samples in 26 batches of 5 samples each for kidney
  2285. biopsies, we were able to train a robust set of fRMA normalization vectors
  2286. that were not meaningfully affected by the random selection of 5 samples
  2287. from each batch.
  2288. As expected, the training process was just as robust for the blood samples
  2289. with 230 samples in 46 batches of 5 samples each.
  2290. Because these vectors were each generated using training samples from a
  2291. single tissue, they are not suitable for general use, unlike the vectors
  2292. provided with fRMA itself.
  2293. They are purpose-build for normalizing a specific type of sample on a specific
  2294. platform.
  2295. \end_layout
  2296. \begin_layout Subsection
  2297. voom
  2298. \end_layout
  2299. \begin_layout Itemize
  2300. Methods like voom designed for RNA-seq can also help with array analysis
  2301. \end_layout
  2302. \begin_layout Itemize
  2303. Extracting and modeling confounders common to many features improves model
  2304. correspondence to known biology
  2305. \end_layout
  2306. \begin_layout Chapter
  2307. Globin-blocking for more effective blood RNA-seq analysis in primate animal
  2308. model
  2309. \end_layout
  2310. \begin_layout Standard
  2311. \begin_inset Flex TODO Note (inline)
  2312. status open
  2313. \begin_layout Plain Layout
  2314. Choose between above and the paper title: Optimizing yield of deep RNA sequencin
  2315. g for gene expression profiling by globin reduction of peripheral blood
  2316. samples from cynomolgus monkeys (Macaca fascicularis).
  2317. \end_layout
  2318. \end_inset
  2319. \end_layout
  2320. \begin_layout Standard
  2321. \begin_inset Flex TODO Note (inline)
  2322. status open
  2323. \begin_layout Plain Layout
  2324. Chapter author list: https://tex.stackexchange.com/questions/156862/displaying-aut
  2325. hor-for-each-chapter-in-book Every chapter gets an author list, which may
  2326. or may not be part of a citation to a published/preprinted paper.
  2327. \end_layout
  2328. \end_inset
  2329. \end_layout
  2330. \begin_layout Standard
  2331. \begin_inset Flex TODO Note (inline)
  2332. status open
  2333. \begin_layout Plain Layout
  2334. Preprint then cite the paper
  2335. \end_layout
  2336. \end_inset
  2337. \end_layout
  2338. \begin_layout Section*
  2339. Abstract
  2340. \end_layout
  2341. \begin_layout Paragraph
  2342. Background
  2343. \end_layout
  2344. \begin_layout Standard
  2345. Primate blood contains high concentrations of globin messenger RNA.
  2346. Globin reduction is a standard technique used to improve the expression
  2347. results obtained by DNA microarrays on RNA from blood samples.
  2348. However, with whole transcriptome RNA-sequencing (RNA-seq) quickly replacing
  2349. microarrays for many applications, the impact of globin reduction for RNA-seq
  2350. has not been previously studied.
  2351. Moreover, no off-the-shelf kits are available for globin reduction in nonhuman
  2352. primates.
  2353. \end_layout
  2354. \begin_layout Paragraph
  2355. Results
  2356. \end_layout
  2357. \begin_layout Standard
  2358. Here we report a protocol for RNA-seq in primate blood samples that uses
  2359. complimentary oligonucleotides to block reverse transcription of the alpha
  2360. and beta globin genes.
  2361. In test samples from cynomolgus monkeys (Macaca fascicularis), this globin
  2362. blocking protocol approximately doubles the yield of informative (non-globin)
  2363. reads by greatly reducing the fraction of globin reads, while also improving
  2364. the consistency in sequencing depth between samples.
  2365. The increased yield enables detection of about 2000 more genes, significantly
  2366. increases the correlation in measured gene expression levels between samples,
  2367. and increases the sensitivity of differential gene expression tests.
  2368. \end_layout
  2369. \begin_layout Paragraph
  2370. Conclusions
  2371. \end_layout
  2372. \begin_layout Standard
  2373. These results show that globin blocking significantly improves the cost-effectiv
  2374. eness of mRNA sequencing in primate blood samples by doubling the yield
  2375. of useful reads, allowing detection of more genes, and improving the precision
  2376. of gene expression measurements.
  2377. Based on these results, a globin reducing or blocking protocol is recommended
  2378. for all RNA-seq studies of primate blood samples.
  2379. \end_layout
  2380. \begin_layout Section
  2381. Approach
  2382. \end_layout
  2383. \begin_layout Standard
  2384. \begin_inset Note Note
  2385. status open
  2386. \begin_layout Plain Layout
  2387. Consider putting some of this in the Intro chapter
  2388. \end_layout
  2389. \begin_layout Itemize
  2390. Cynomolgus monkeys as a model organism
  2391. \end_layout
  2392. \begin_deeper
  2393. \begin_layout Itemize
  2394. Highly related to humans
  2395. \end_layout
  2396. \begin_layout Itemize
  2397. Small size and short life cycle - good research animal
  2398. \end_layout
  2399. \begin_layout Itemize
  2400. Genomics resources still in development
  2401. \end_layout
  2402. \end_deeper
  2403. \begin_layout Itemize
  2404. Inadequacy of existing blood RNA-seq protocols
  2405. \end_layout
  2406. \begin_deeper
  2407. \begin_layout Itemize
  2408. Existing protocols use a separate globin pulldown step, slowing down processing
  2409. \end_layout
  2410. \end_deeper
  2411. \end_inset
  2412. \end_layout
  2413. \begin_layout Standard
  2414. Increasingly, researchers are turning to high-throughput mRNA sequencing
  2415. technologies (RNA-seq) in preference to expression microarrays for analysis
  2416. of gene expression
  2417. \begin_inset CommandInset citation
  2418. LatexCommand cite
  2419. key "Mutz2012"
  2420. literal "false"
  2421. \end_inset
  2422. .
  2423. The advantages are even greater for study of model organisms with no well-estab
  2424. lished array platforms available, such as the cynomolgus monkey (Macaca
  2425. fascicularis).
  2426. High fractions of globin mRNA are naturally present in mammalian peripheral
  2427. blood samples (up to 70% of total mRNA) and these are known to interfere
  2428. with the results of array-based expression profiling
  2429. \begin_inset CommandInset citation
  2430. LatexCommand cite
  2431. key "Winn2010"
  2432. literal "false"
  2433. \end_inset
  2434. .
  2435. The importance of globin reduction for RNA-seq of blood has only been evaluated
  2436. for a deepSAGE protocol on human samples
  2437. \begin_inset CommandInset citation
  2438. LatexCommand cite
  2439. key "Mastrokolias2012"
  2440. literal "false"
  2441. \end_inset
  2442. .
  2443. In the present report, we evaluated globin reduction using custom blocking
  2444. oligonucleotides for deep RNA-seq of peripheral blood samples from a nonhuman
  2445. primate, cynomolgus monkey, using the Illumina technology platform.
  2446. We demonstrate that globin reduction significantly improves the cost-effectiven
  2447. ess of RNA-seq in blood samples.
  2448. Thus, our protocol offers a significant advantage to any investigator planning
  2449. to use RNA-seq for gene expression profiling of nonhuman primate blood
  2450. samples.
  2451. Our method can be generally applied to any species by designing complementary
  2452. oligonucleotide blocking probes to the globin gene sequences of that species.
  2453. Indeed, any highly expressed but biologically uninformative transcripts
  2454. can also be blocked to further increase sequencing efficiency and value
  2455. \begin_inset CommandInset citation
  2456. LatexCommand cite
  2457. key "Arnaud2016"
  2458. literal "false"
  2459. \end_inset
  2460. .
  2461. \end_layout
  2462. \begin_layout Section
  2463. Methods
  2464. \end_layout
  2465. \begin_layout Subsection*
  2466. Sample collection
  2467. \end_layout
  2468. \begin_layout Standard
  2469. All research reported here was done under IACUC-approved protocols at the
  2470. University of Miami and complied with all applicable federal and state
  2471. regulations and ethical principles for nonhuman primate research.
  2472. Blood draws occurred between 16 April 2012 and 18 June 2015.
  2473. The experimental system involved intrahepatic pancreatic islet transplantation
  2474. into Cynomolgus monkeys with induced diabetes mellitus with or without
  2475. concomitant infusion of mesenchymal stem cells.
  2476. Blood was collected at serial time points before and after transplantation
  2477. into PAXgene Blood RNA tubes (PreAnalytiX/Qiagen, Valencia, CA) at the
  2478. precise volume:volume ratio of 2.5 ml whole blood into 6.9 ml of PAX gene
  2479. additive.
  2480. \end_layout
  2481. \begin_layout Subsection*
  2482. Globin Blocking
  2483. \end_layout
  2484. \begin_layout Standard
  2485. Four oligonucleotides were designed to hybridize to the 3’ end of the transcript
  2486. s for Cynomolgus HBA1, HBA2 and HBB, with two hybridization sites for HBB
  2487. and 2 sites for HBA (the chosen sites were identical in both HBA genes).
  2488. All oligos were purchased from Sigma and were entirely composed of 2’O-Me
  2489. bases with a C3 spacer positioned at the 3’ ends to prevent any polymerase
  2490. mediated primer extension.
  2491. \end_layout
  2492. \begin_layout Quote
  2493. HBA1/2 site 1: GCCCACUCAGACUUUAUUCAAAG-C3spacer
  2494. \end_layout
  2495. \begin_layout Quote
  2496. HBA1/2 site 2: GGUGCAAGGAGGGGAGGAG-C3spacer
  2497. \end_layout
  2498. \begin_layout Quote
  2499. HBB site 1: AAUGAAAAUAAAUGUUUUUUAUUAG-C3spacer
  2500. \end_layout
  2501. \begin_layout Quote
  2502. HBB site 2: CUCAAGGCCCUUCAUAAUAUCCC-C3spacer
  2503. \end_layout
  2504. \begin_layout Subsection*
  2505. RNA-seq Library Preparation
  2506. \end_layout
  2507. \begin_layout Standard
  2508. Sequencing libraries were prepared with 200ng total RNA from each sample.
  2509. Polyadenylated mRNA was selected from 200 ng aliquots of cynomologus blood-deri
  2510. ved total RNA using Ambion Dynabeads Oligo(dT)25 beads (Invitrogen) following
  2511. manufacturer’s recommended protocol.
  2512. PolyA selected RNA was then combined with 8 pmol of HBA1/2 (site 1), 8
  2513. pmol of HBA1/2 (site 2), 12 pmol of HBB (site 1) and 12 pmol of HBB (site
  2514. 2) oligonucleotides.
  2515. In addition, 20 pmol of RT primer containing a portion of the Illumina
  2516. adapter sequence (B-oligo-dTV: GAGTTCCTTGGCACCCGAGAATTCCATTTTTTTTTTTTTTTTTTTV)
  2517. and 4 µL of 5X First Strand buffer (250 mM Tris-HCl pH 8.3, 375 mM KCl,
  2518. 15mM MgCl2) were added in a total volume of 15 µL.
  2519. The RNA was fragmented by heating this cocktail for 3 minutes at 95°C and
  2520. then placed on ice.
  2521. This was followed by the addition of 2 µL 0.1 M DTT, 1 µL RNaseOUT, 1 µL
  2522. 10mM dNTPs 10% biotin-16 aminoallyl-2’- dUTP and 10% biotin-16 aminoallyl-2’-
  2523. dCTP (TriLink Biotech, San Diego, CA), 1 µL Superscript II (200U/ µL, Thermo-Fi
  2524. sher).
  2525. A second “unblocked” library was prepared in the same way for each sample
  2526. but replacing the blocking oligos with an equivalent volume of water.
  2527. The reaction was carried out at 25°C for 15 minutes and 42°C for 40 minutes,
  2528. followed by incubation at 75°C for 10 minutes to inactivate the reverse
  2529. transcriptase.
  2530. \end_layout
  2531. \begin_layout Standard
  2532. The cDNA/RNA hybrid molecules were purified using 1.8X Ampure XP beads (Agencourt
  2533. ) following supplier’s recommended protocol.
  2534. The cDNA/RNA hybrid was eluted in 25 µL of 10 mM Tris-HCl pH 8.0, and then
  2535. bound to 25 µL of M280 Magnetic Streptavidin beads washed per recommended
  2536. protocol (Thermo-Fisher).
  2537. After 30 minutes of binding, beads were washed one time in 100 µL 0.1N NaOH
  2538. to denature and remove the bound RNA, followed by two 100 µL washes with
  2539. 1X TE buffer.
  2540. \end_layout
  2541. \begin_layout Standard
  2542. Subsequent attachment of the 5-prime Illumina A adapter was performed by
  2543. on-bead random primer extension of the following sequence (A-N8 primer:
  2544. TTCAGAGTTCTACAGTCCGACGATCNNNNNNNN).
  2545. Briefly, beads were resuspended in a 20 µL reaction containing 5 µM A-N8
  2546. primer, 40mM Tris-HCl pH 7.5, 20mM MgCl2, 50mM NaCl, 0.325U/µL Sequenase
  2547. 2.0 (Affymetrix, Santa Clara, CA), 0.0025U/µL inorganic pyrophosphatase (Affymetr
  2548. ix) and 300 µM each dNTP.
  2549. Reaction was incubated at 22°C for 30 minutes, then beads were washed 2
  2550. times with 1X TE buffer (200µL).
  2551. \end_layout
  2552. \begin_layout Standard
  2553. The magnetic streptavidin beads were resuspended in 34 µL nuclease-free
  2554. water and added directly to a PCR tube.
  2555. The two Illumina protocol-specified PCR primers were added at 0.53 µM (Illumina
  2556. TruSeq Universal Primer 1 and Illumina TruSeq barcoded PCR primer 2), along
  2557. with 40 µL 2X KAPA HiFi Hotstart ReadyMix (KAPA, Willmington MA) and thermocycl
  2558. ed as follows: starting with 98°C (2 min-hold); 15 cycles of 98°C, 20sec;
  2559. 60°C, 30sec; 72°C, 30sec; and finished with a 72°C (2 min-hold).
  2560. \end_layout
  2561. \begin_layout Standard
  2562. PCR products were purified with 1X Ampure Beads following manufacturer’s
  2563. recommended protocol.
  2564. Libraries were then analyzed using the Agilent TapeStation and quantitation
  2565. of desired size range was performed by “smear analysis”.
  2566. Samples were pooled in equimolar batches of 16 samples.
  2567. Pooled libraries were size selected on 2% agarose gels (E-Gel EX Agarose
  2568. Gels; Thermo-Fisher).
  2569. Products were cut between 250 and 350 bp (corresponding to insert sizes
  2570. of 130 to 230 bps).
  2571. Finished library pools were then sequenced on the Illumina NextSeq500 instrumen
  2572. t with 75 base read lengths.
  2573. \end_layout
  2574. \begin_layout Subsection*
  2575. Read alignment and counting
  2576. \end_layout
  2577. \begin_layout Standard
  2578. Reads were aligned to the cynomolgus genome using STAR
  2579. \begin_inset CommandInset citation
  2580. LatexCommand cite
  2581. key "Dobin2013,Wilson2013"
  2582. literal "false"
  2583. \end_inset
  2584. .
  2585. Counts of uniquely mapped reads were obtained for every gene in each sample
  2586. with the “featureCounts” function from the Rsubread package, using each
  2587. of the three possibilities for the “strandSpecific” option: sense, antisense,
  2588. and unstranded
  2589. \begin_inset CommandInset citation
  2590. LatexCommand cite
  2591. key "Liao2014"
  2592. literal "false"
  2593. \end_inset
  2594. .
  2595. A few artifacts in the cynomolgus genome annotation complicated read counting.
  2596. First, no ortholog is annotated for alpha globin in the cynomolgus genome,
  2597. presumably because the human genome has two alpha globin genes with nearly
  2598. identical sequences, making the orthology relationship ambiguous.
  2599. However, two loci in the cynomolgus genome are as “hemoglobin subunit alpha-lik
  2600. e” (LOC102136192 and LOC102136846).
  2601. LOC102136192 is annotated as a pseudogene while LOC102136846 is annotated
  2602. as protein-coding.
  2603. Our globin reduction protocol was designed to include blocking of these
  2604. two genes.
  2605. Indeed, these two genes have almost the same read counts in each library
  2606. as the properly-annotated HBB gene and much larger counts than any other
  2607. gene in the unblocked libraries, giving confidence that reads derived from
  2608. the real alpha globin are mapping to both genes.
  2609. Thus, reads from both of these loci were counted as alpha globin reads
  2610. in all further analyses.
  2611. The second artifact is a small, uncharacterized non-coding RNA gene (LOC1021365
  2612. 91), which overlaps the HBA-like gene (LOC102136192) on the opposite strand.
  2613. If counting is not performed in stranded mode (or if a non-strand-specific
  2614. sequencing protocol is used), many reads mapping to the globin gene will
  2615. be discarded as ambiguous due to their overlap with this ncRNA gene, resulting
  2616. in significant undercounting of globin reads.
  2617. Therefore, stranded sense counts were used for all further analysis in
  2618. the present study to insure that we accurately accounted for globin transcript
  2619. reduction.
  2620. However, we note that stranded reads are not necessary for RNA-seq using
  2621. our protocol in standard practice.
  2622. \end_layout
  2623. \begin_layout Subsection*
  2624. Normalization and Exploratory Data Analysis
  2625. \end_layout
  2626. \begin_layout Standard
  2627. Libraries were normalized by computing scaling factors using the edgeR package’s
  2628. Trimmed Mean of M-values method
  2629. \begin_inset CommandInset citation
  2630. LatexCommand cite
  2631. key "Robinson2010"
  2632. literal "false"
  2633. \end_inset
  2634. .
  2635. Log2 counts per million values (logCPM) were calculated using the cpm function
  2636. in edgeR for individual samples and aveLogCPM function for averages across
  2637. groups of samples, using those functions’ default prior count values to
  2638. avoid taking the logarithm of 0.
  2639. Genes were considered “present” if their average normalized logCPM values
  2640. across all libraries were at least -1.
  2641. Normalizing for gene length was unnecessary because the sequencing protocol
  2642. is 3’-biased and hence the expected read count for each gene is related
  2643. to the transcript’s copy number but not its length.
  2644. \end_layout
  2645. \begin_layout Standard
  2646. In order to assess the effect of blocking on reproducibility, Pearson and
  2647. Spearman correlation coefficients were computed between the logCPM values
  2648. for every pair of libraries within the globin-blocked (GB) and unblocked
  2649. (non-GB) groups, and edgeR's “estimateDisp” function was used to compute
  2650. negative binomial dispersions separately for the two groups
  2651. \begin_inset CommandInset citation
  2652. LatexCommand cite
  2653. key "Chen2014"
  2654. literal "false"
  2655. \end_inset
  2656. .
  2657. \end_layout
  2658. \begin_layout Subsection*
  2659. Differential Expression Analysis
  2660. \end_layout
  2661. \begin_layout Standard
  2662. All tests for differential gene expression were performed using edgeR, by
  2663. first fitting a negative binomial generalized linear model to the counts
  2664. and normalization factors and then performing a quasi-likelihood F-test
  2665. with robust estimation of outlier gene dispersions
  2666. \begin_inset CommandInset citation
  2667. LatexCommand cite
  2668. key "Lund2012,Phipson2016"
  2669. literal "false"
  2670. \end_inset
  2671. .
  2672. To investigate the effects of globin blocking on each gene, an additive
  2673. model was fit to the full data with coefficients for globin blocking and
  2674. SampleID.
  2675. To test the effect of globin blocking on detection of differentially expressed
  2676. genes, the GB samples and non-GB samples were each analyzed independently
  2677. as follows: for each animal with both a pre-transplant and a post-transplant
  2678. time point in the data set, the pre-transplant sample and the earliest
  2679. post-transplant sample were selected, and all others were excluded, yielding
  2680. a pre-/post-transplant pair of samples for each animal (N=7 animals with
  2681. paired samples).
  2682. These samples were analyzed for pre-transplant vs.
  2683. post-transplant differential gene expression while controlling for inter-animal
  2684. variation using an additive model with coefficients for transplant and
  2685. animal ID.
  2686. In all analyses, p-values were adjusted using the Benjamini-Hochberg procedure
  2687. for FDR correction
  2688. \begin_inset CommandInset citation
  2689. LatexCommand cite
  2690. key "Benjamini1995"
  2691. literal "false"
  2692. \end_inset
  2693. .
  2694. \end_layout
  2695. \begin_layout Standard
  2696. \begin_inset Note Note
  2697. status open
  2698. \begin_layout Itemize
  2699. New blood RNA-seq protocol to block reverse transcription of globin genes
  2700. \end_layout
  2701. \begin_layout Itemize
  2702. Blood RNA-seq time course after transplants with/without MSC infusion
  2703. \end_layout
  2704. \end_inset
  2705. \end_layout
  2706. \begin_layout Section
  2707. Results
  2708. \end_layout
  2709. \begin_layout Subsection*
  2710. Globin blocking yields a larger and more consistent fraction of useful reads
  2711. \end_layout
  2712. \begin_layout Standard
  2713. The objective of the present study was to validate a new protocol for deep
  2714. RNA-seq of whole blood drawn into PaxGene tubes from cynomolgus monkeys
  2715. undergoing islet transplantation, with particular focus on minimizing the
  2716. loss of useful sequencing space to uninformative globin reads.
  2717. The details of the analysis with respect to transplant outcomes and the
  2718. impact of mesenchymal stem cell treatment will be reported in a separate
  2719. manuscript (in preparation).
  2720. To focus on the efficacy of our globin blocking protocol, 37 blood samples,
  2721. 16 from pre-transplant and 21 from post-transplant time points, were each
  2722. prepped once with and once without globin blocking oligos, and were then
  2723. sequenced on an Illumina NextSeq500 instrument.
  2724. The number of reads aligning to each gene in the cynomolgus genome was
  2725. counted.
  2726. Table 1 summarizes the distribution of read fractions among the GB and
  2727. non-GB libraries.
  2728. In the libraries with no globin blocking, globin reads made up an average
  2729. of 44.6% of total input reads, while reads assigned to all other genes made
  2730. up an average of 26.3%.
  2731. The remaining reads either aligned to intergenic regions (that include
  2732. long non-coding RNAs) or did not align with any annotated transcripts in
  2733. the current build of the cynomolgus genome.
  2734. In the GB libraries, globin reads made up only 3.48% and reads assigned
  2735. to all other genes increased to 50.4%.
  2736. Thus, globin blocking resulted in a 92.2% reduction in globin reads and
  2737. a 91.6% increase in yield of useful non-globin reads.
  2738. \end_layout
  2739. \begin_layout Standard
  2740. This reduction is not quite as efficient as the previous analysis showed
  2741. for human samples by DeepSAGE (<0.4% globin reads after globin reduction)
  2742. \begin_inset CommandInset citation
  2743. LatexCommand cite
  2744. key "Mastrokolias2012"
  2745. literal "false"
  2746. \end_inset
  2747. .
  2748. Nonetheless, this degree of globin reduction is sufficient to nearly double
  2749. the yield of useful reads.
  2750. Thus, globin blocking cuts the required sequencing effort (and costs) to
  2751. achieve a target coverage depth by almost 50%.
  2752. Consistent with this near doubling of yield, the average difference in
  2753. un-normalized logCPM across all genes between the GB libraries and non-GB
  2754. libraries is approximately 1 (mean = 1.01, median = 1.08), an overall 2-fold
  2755. increase.
  2756. Un-normalized values are used here because the TMM normalization correctly
  2757. identifies this 2-fold difference as biologically irrelevant and removes
  2758. it.
  2759. \end_layout
  2760. \begin_layout Standard
  2761. \begin_inset Float figure
  2762. wide false
  2763. sideways false
  2764. status open
  2765. \begin_layout Plain Layout
  2766. \align center
  2767. \begin_inset Graphics
  2768. filename graphics/Globin Paper/figure1 - globin-fractions.pdf
  2769. \end_inset
  2770. \end_layout
  2771. \begin_layout Plain Layout
  2772. \begin_inset Caption Standard
  2773. \begin_layout Plain Layout
  2774. \series bold
  2775. \begin_inset Argument 1
  2776. status collapsed
  2777. \begin_layout Plain Layout
  2778. Fraction of genic reads in each sample aligned to non-globin genes, with
  2779. and without globin blocking (GB).
  2780. \end_layout
  2781. \end_inset
  2782. \begin_inset CommandInset label
  2783. LatexCommand label
  2784. name "fig:Fraction-of-genic-reads"
  2785. \end_inset
  2786. Fraction of genic reads in each sample aligned to non-globin genes, with
  2787. and without globin blocking (GB).
  2788. \series default
  2789. All reads in each sequencing library were aligned to the cyno genome, and
  2790. the number of reads uniquely aligning to each gene was counted.
  2791. For each sample, counts were summed separately for all globin genes and
  2792. for the remainder of the genes (non-globin genes), and the fraction of
  2793. genic reads aligned to non-globin genes was computed.
  2794. Each point represents an individual sample.
  2795. Gray + signs indicate the means for globin-blocked libraries and unblocked
  2796. libraries.
  2797. The overall distribution for each group is represented as a notched box
  2798. plots.
  2799. Points are randomly spread vertically to avoid excessive overlapping.
  2800. \end_layout
  2801. \end_inset
  2802. \end_layout
  2803. \begin_layout Plain Layout
  2804. \end_layout
  2805. \end_inset
  2806. \end_layout
  2807. \begin_layout Standard
  2808. \begin_inset Float table
  2809. placement p
  2810. wide false
  2811. sideways true
  2812. status open
  2813. \begin_layout Plain Layout
  2814. \align center
  2815. \begin_inset Tabular
  2816. <lyxtabular version="3" rows="4" columns="7">
  2817. <features tabularvalignment="middle">
  2818. <column alignment="center" valignment="top">
  2819. <column alignment="center" valignment="top">
  2820. <column alignment="center" valignment="top">
  2821. <column alignment="center" valignment="top">
  2822. <column alignment="center" valignment="top">
  2823. <column alignment="center" valignment="top">
  2824. <column alignment="center" valignment="top">
  2825. <row>
  2826. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2827. \begin_inset Text
  2828. \begin_layout Plain Layout
  2829. \end_layout
  2830. \end_inset
  2831. </cell>
  2832. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2833. \begin_inset Text
  2834. \begin_layout Plain Layout
  2835. \family roman
  2836. \series medium
  2837. \shape up
  2838. \size normal
  2839. \emph off
  2840. \bar no
  2841. \strikeout off
  2842. \xout off
  2843. \uuline off
  2844. \uwave off
  2845. \noun off
  2846. \color none
  2847. Percent of Total Reads
  2848. \end_layout
  2849. \end_inset
  2850. </cell>
  2851. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2852. \begin_inset Text
  2853. \begin_layout Plain Layout
  2854. \end_layout
  2855. \end_inset
  2856. </cell>
  2857. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2858. \begin_inset Text
  2859. \begin_layout Plain Layout
  2860. \end_layout
  2861. \end_inset
  2862. </cell>
  2863. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  2864. \begin_inset Text
  2865. \begin_layout Plain Layout
  2866. \end_layout
  2867. \end_inset
  2868. </cell>
  2869. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2870. \begin_inset Text
  2871. \begin_layout Plain Layout
  2872. \family roman
  2873. \series medium
  2874. \shape up
  2875. \size normal
  2876. \emph off
  2877. \bar no
  2878. \strikeout off
  2879. \xout off
  2880. \uuline off
  2881. \uwave off
  2882. \noun off
  2883. \color none
  2884. Percent of Genic Reads
  2885. \end_layout
  2886. \end_inset
  2887. </cell>
  2888. <cell multicolumn="2" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  2889. \begin_inset Text
  2890. \begin_layout Plain Layout
  2891. \end_layout
  2892. \end_inset
  2893. </cell>
  2894. </row>
  2895. <row>
  2896. <cell alignment="center" valignment="top" bottomline="true" leftline="true" usebox="none">
  2897. \begin_inset Text
  2898. \begin_layout Plain Layout
  2899. GB
  2900. \end_layout
  2901. \end_inset
  2902. </cell>
  2903. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2904. \begin_inset Text
  2905. \begin_layout Plain Layout
  2906. \family roman
  2907. \series medium
  2908. \shape up
  2909. \size normal
  2910. \emph off
  2911. \bar no
  2912. \strikeout off
  2913. \xout off
  2914. \uuline off
  2915. \uwave off
  2916. \noun off
  2917. \color none
  2918. Non-globin Reads
  2919. \end_layout
  2920. \end_inset
  2921. </cell>
  2922. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2923. \begin_inset Text
  2924. \begin_layout Plain Layout
  2925. \family roman
  2926. \series medium
  2927. \shape up
  2928. \size normal
  2929. \emph off
  2930. \bar no
  2931. \strikeout off
  2932. \xout off
  2933. \uuline off
  2934. \uwave off
  2935. \noun off
  2936. \color none
  2937. Globin Reads
  2938. \end_layout
  2939. \end_inset
  2940. </cell>
  2941. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2942. \begin_inset Text
  2943. \begin_layout Plain Layout
  2944. \family roman
  2945. \series medium
  2946. \shape up
  2947. \size normal
  2948. \emph off
  2949. \bar no
  2950. \strikeout off
  2951. \xout off
  2952. \uuline off
  2953. \uwave off
  2954. \noun off
  2955. \color none
  2956. All Genic Reads
  2957. \end_layout
  2958. \end_inset
  2959. </cell>
  2960. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2961. \begin_inset Text
  2962. \begin_layout Plain Layout
  2963. \family roman
  2964. \series medium
  2965. \shape up
  2966. \size normal
  2967. \emph off
  2968. \bar no
  2969. \strikeout off
  2970. \xout off
  2971. \uuline off
  2972. \uwave off
  2973. \noun off
  2974. \color none
  2975. All Aligned Reads
  2976. \end_layout
  2977. \end_inset
  2978. </cell>
  2979. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  2980. \begin_inset Text
  2981. \begin_layout Plain Layout
  2982. \family roman
  2983. \series medium
  2984. \shape up
  2985. \size normal
  2986. \emph off
  2987. \bar no
  2988. \strikeout off
  2989. \xout off
  2990. \uuline off
  2991. \uwave off
  2992. \noun off
  2993. \color none
  2994. Non-globin Reads
  2995. \end_layout
  2996. \end_inset
  2997. </cell>
  2998. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  2999. \begin_inset Text
  3000. \begin_layout Plain Layout
  3001. \family roman
  3002. \series medium
  3003. \shape up
  3004. \size normal
  3005. \emph off
  3006. \bar no
  3007. \strikeout off
  3008. \xout off
  3009. \uuline off
  3010. \uwave off
  3011. \noun off
  3012. \color none
  3013. Globin Reads
  3014. \end_layout
  3015. \end_inset
  3016. </cell>
  3017. </row>
  3018. <row>
  3019. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3020. \begin_inset Text
  3021. \begin_layout Plain Layout
  3022. \family roman
  3023. \series medium
  3024. \shape up
  3025. \size normal
  3026. \emph off
  3027. \bar no
  3028. \strikeout off
  3029. \xout off
  3030. \uuline off
  3031. \uwave off
  3032. \noun off
  3033. \color none
  3034. Yes
  3035. \end_layout
  3036. \end_inset
  3037. </cell>
  3038. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3039. \begin_inset Text
  3040. \begin_layout Plain Layout
  3041. \family roman
  3042. \series medium
  3043. \shape up
  3044. \size normal
  3045. \emph off
  3046. \bar no
  3047. \strikeout off
  3048. \xout off
  3049. \uuline off
  3050. \uwave off
  3051. \noun off
  3052. \color none
  3053. 50.4% ± 6.82
  3054. \end_layout
  3055. \end_inset
  3056. </cell>
  3057. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3058. \begin_inset Text
  3059. \begin_layout Plain Layout
  3060. \family roman
  3061. \series medium
  3062. \shape up
  3063. \size normal
  3064. \emph off
  3065. \bar no
  3066. \strikeout off
  3067. \xout off
  3068. \uuline off
  3069. \uwave off
  3070. \noun off
  3071. \color none
  3072. 3.48% ± 2.94
  3073. \end_layout
  3074. \end_inset
  3075. </cell>
  3076. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3077. \begin_inset Text
  3078. \begin_layout Plain Layout
  3079. \family roman
  3080. \series medium
  3081. \shape up
  3082. \size normal
  3083. \emph off
  3084. \bar no
  3085. \strikeout off
  3086. \xout off
  3087. \uuline off
  3088. \uwave off
  3089. \noun off
  3090. \color none
  3091. 53.9% ± 6.81
  3092. \end_layout
  3093. \end_inset
  3094. </cell>
  3095. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3096. \begin_inset Text
  3097. \begin_layout Plain Layout
  3098. \family roman
  3099. \series medium
  3100. \shape up
  3101. \size normal
  3102. \emph off
  3103. \bar no
  3104. \strikeout off
  3105. \xout off
  3106. \uuline off
  3107. \uwave off
  3108. \noun off
  3109. \color none
  3110. 89.7% ± 2.40
  3111. \end_layout
  3112. \end_inset
  3113. </cell>
  3114. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3115. \begin_inset Text
  3116. \begin_layout Plain Layout
  3117. \family roman
  3118. \series medium
  3119. \shape up
  3120. \size normal
  3121. \emph off
  3122. \bar no
  3123. \strikeout off
  3124. \xout off
  3125. \uuline off
  3126. \uwave off
  3127. \noun off
  3128. \color none
  3129. 93.5% ± 5.25
  3130. \end_layout
  3131. \end_inset
  3132. </cell>
  3133. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3134. \begin_inset Text
  3135. \begin_layout Plain Layout
  3136. \family roman
  3137. \series medium
  3138. \shape up
  3139. \size normal
  3140. \emph off
  3141. \bar no
  3142. \strikeout off
  3143. \xout off
  3144. \uuline off
  3145. \uwave off
  3146. \noun off
  3147. \color none
  3148. 6.49% ± 5.25
  3149. \end_layout
  3150. \end_inset
  3151. </cell>
  3152. </row>
  3153. <row>
  3154. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3155. \begin_inset Text
  3156. \begin_layout Plain Layout
  3157. \family roman
  3158. \series medium
  3159. \shape up
  3160. \size normal
  3161. \emph off
  3162. \bar no
  3163. \strikeout off
  3164. \xout off
  3165. \uuline off
  3166. \uwave off
  3167. \noun off
  3168. \color none
  3169. No
  3170. \end_layout
  3171. \end_inset
  3172. </cell>
  3173. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3174. \begin_inset Text
  3175. \begin_layout Plain Layout
  3176. \family roman
  3177. \series medium
  3178. \shape up
  3179. \size normal
  3180. \emph off
  3181. \bar no
  3182. \strikeout off
  3183. \xout off
  3184. \uuline off
  3185. \uwave off
  3186. \noun off
  3187. \color none
  3188. 26.3% ± 8.95
  3189. \end_layout
  3190. \end_inset
  3191. </cell>
  3192. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3193. \begin_inset Text
  3194. \begin_layout Plain Layout
  3195. \family roman
  3196. \series medium
  3197. \shape up
  3198. \size normal
  3199. \emph off
  3200. \bar no
  3201. \strikeout off
  3202. \xout off
  3203. \uuline off
  3204. \uwave off
  3205. \noun off
  3206. \color none
  3207. 44.6% ± 16.6
  3208. \end_layout
  3209. \end_inset
  3210. </cell>
  3211. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3212. \begin_inset Text
  3213. \begin_layout Plain Layout
  3214. \family roman
  3215. \series medium
  3216. \shape up
  3217. \size normal
  3218. \emph off
  3219. \bar no
  3220. \strikeout off
  3221. \xout off
  3222. \uuline off
  3223. \uwave off
  3224. \noun off
  3225. \color none
  3226. 70.1% ± 9.38
  3227. \end_layout
  3228. \end_inset
  3229. </cell>
  3230. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3231. \begin_inset Text
  3232. \begin_layout Plain Layout
  3233. \family roman
  3234. \series medium
  3235. \shape up
  3236. \size normal
  3237. \emph off
  3238. \bar no
  3239. \strikeout off
  3240. \xout off
  3241. \uuline off
  3242. \uwave off
  3243. \noun off
  3244. \color none
  3245. 90.7% ± 5.16
  3246. \end_layout
  3247. \end_inset
  3248. </cell>
  3249. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3250. \begin_inset Text
  3251. \begin_layout Plain Layout
  3252. \family roman
  3253. \series medium
  3254. \shape up
  3255. \size normal
  3256. \emph off
  3257. \bar no
  3258. \strikeout off
  3259. \xout off
  3260. \uuline off
  3261. \uwave off
  3262. \noun off
  3263. \color none
  3264. 38.8% ± 17.1
  3265. \end_layout
  3266. \end_inset
  3267. </cell>
  3268. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3269. \begin_inset Text
  3270. \begin_layout Plain Layout
  3271. \family roman
  3272. \series medium
  3273. \shape up
  3274. \size normal
  3275. \emph off
  3276. \bar no
  3277. \strikeout off
  3278. \xout off
  3279. \uuline off
  3280. \uwave off
  3281. \noun off
  3282. \color none
  3283. 61.2% ± 17.1
  3284. \end_layout
  3285. \end_inset
  3286. </cell>
  3287. </row>
  3288. </lyxtabular>
  3289. \end_inset
  3290. \end_layout
  3291. \begin_layout Plain Layout
  3292. \begin_inset Caption Standard
  3293. \begin_layout Plain Layout
  3294. \series bold
  3295. \begin_inset Argument 1
  3296. status collapsed
  3297. \begin_layout Plain Layout
  3298. Fractions of reads mapping to genomic features in GB and non-GB samples.
  3299. \end_layout
  3300. \end_inset
  3301. \begin_inset CommandInset label
  3302. LatexCommand label
  3303. name "tab:Fractions-of-reads"
  3304. \end_inset
  3305. Fractions of reads mapping to genomic features in GB and non-GB samples.
  3306. \series default
  3307. All values are given as mean ± standard deviation.
  3308. \end_layout
  3309. \end_inset
  3310. \end_layout
  3311. \begin_layout Plain Layout
  3312. \end_layout
  3313. \end_inset
  3314. \end_layout
  3315. \begin_layout Standard
  3316. Another important aspect is that the standard deviations in Table
  3317. \begin_inset CommandInset ref
  3318. LatexCommand ref
  3319. reference "tab:Fractions-of-reads"
  3320. plural "false"
  3321. caps "false"
  3322. noprefix "false"
  3323. \end_inset
  3324. are uniformly smaller in the GB samples than the non-GB ones, indicating
  3325. much greater consistency of yield.
  3326. This is best seen in the percentage of non-globin reads as a fraction of
  3327. total reads aligned to annotated genes (genic reads).
  3328. For the non-GB samples, this measure ranges from 10.9% to 80.9%, while for
  3329. the GB samples it ranges from 81.9% to 99.9% (Figure
  3330. \begin_inset CommandInset ref
  3331. LatexCommand ref
  3332. reference "fig:Fraction-of-genic-reads"
  3333. plural "false"
  3334. caps "false"
  3335. noprefix "false"
  3336. \end_inset
  3337. ).
  3338. This means that for applications where it is critical that each sample
  3339. achieve a specified minimum coverage in order to provide useful information,
  3340. it would be necessary to budget up to 10 times the sequencing depth per
  3341. sample without globin blocking, even though the average yield improvement
  3342. for globin blocking is only 2-fold, because every sample has a chance of
  3343. being 90% globin and 10% useful reads.
  3344. Hence, the more consistent behavior of GB samples makes planning an experiment
  3345. easier and more efficient because it eliminates the need to over-sequence
  3346. every sample in order to guard against the worst case of a high-globin
  3347. fraction.
  3348. \end_layout
  3349. \begin_layout Subsection*
  3350. Globin blocking lowers the noise floor and allows detection of about 2000
  3351. more genes
  3352. \end_layout
  3353. \begin_layout Standard
  3354. \begin_inset Flex TODO Note (inline)
  3355. status open
  3356. \begin_layout Plain Layout
  3357. Remove redundant titles from figures
  3358. \end_layout
  3359. \end_inset
  3360. \end_layout
  3361. \begin_layout Standard
  3362. \begin_inset Float figure
  3363. wide false
  3364. sideways false
  3365. status open
  3366. \begin_layout Plain Layout
  3367. \align center
  3368. \begin_inset Graphics
  3369. filename graphics/Globin Paper/figure2 - aveLogCPM-colored.pdf
  3370. \end_inset
  3371. \end_layout
  3372. \begin_layout Plain Layout
  3373. \begin_inset Caption Standard
  3374. \begin_layout Plain Layout
  3375. \series bold
  3376. \begin_inset Argument 1
  3377. status collapsed
  3378. \begin_layout Plain Layout
  3379. Distributions of average group gene abundances when normalized separately
  3380. or together.
  3381. \end_layout
  3382. \end_inset
  3383. \begin_inset CommandInset label
  3384. LatexCommand label
  3385. name "fig:logcpm-dists"
  3386. \end_inset
  3387. Distributions of average group gene abundances when normalized separately
  3388. or together.
  3389. \series default
  3390. All reads in each sequencing library were aligned to the cyno genome, and
  3391. the number of reads uniquely aligning to each gene was counted.
  3392. Genes with zero counts in all libraries were discarded.
  3393. Libraries were normalized using the TMM method.
  3394. Libraries were split into globin-blocked (GB) and non-GB groups and the
  3395. average abundance for each gene in both groups, measured in log2 counts
  3396. per million reads counted, was computed using the aveLogCPM function.
  3397. The distribution of average gene logCPM values was plotted for both groups
  3398. using a kernel density plot to approximate a continuous distribution.
  3399. The logCPM GB distributions are marked in red, non-GB in blue.
  3400. The black vertical line denotes the chosen detection threshold of -1.
  3401. Top panel: Libraries were split into GB and non-GB groups first and normalized
  3402. separately.
  3403. Bottom panel: Libraries were all normalized together first and then split
  3404. into groups.
  3405. \end_layout
  3406. \end_inset
  3407. \end_layout
  3408. \begin_layout Plain Layout
  3409. \end_layout
  3410. \end_inset
  3411. \end_layout
  3412. \begin_layout Standard
  3413. Since globin blocking yields more usable sequencing depth, it should also
  3414. allow detection of more genes at any given threshold.
  3415. When we looked at the distribution of average normalized logCPM values
  3416. across all libraries for genes with at least one read assigned to them,
  3417. we observed the expected bimodal distribution, with a high-abundance "signal"
  3418. peak representing detected genes and a low-abundance "noise" peak representing
  3419. genes whose read count did not rise above the noise floor (Figure
  3420. \begin_inset CommandInset ref
  3421. LatexCommand ref
  3422. reference "fig:logcpm-dists"
  3423. plural "false"
  3424. caps "false"
  3425. noprefix "false"
  3426. \end_inset
  3427. ).
  3428. Consistent with the 2-fold increase in raw counts assigned to non-globin
  3429. genes, the signal peak for GB samples is shifted to the right relative
  3430. to the non-GB signal peak.
  3431. When all the samples are normalized together, this difference is normalized
  3432. out, lining up the signal peaks, and this reveals that, as expected, the
  3433. noise floor for the GB samples is about 2-fold lower.
  3434. This greater separation between signal and noise peaks in the GB samples
  3435. means that low-expression genes should be more easily detected and more
  3436. precisely quantified than in the non-GB samples.
  3437. \end_layout
  3438. \begin_layout Standard
  3439. \begin_inset Float figure
  3440. wide false
  3441. sideways false
  3442. status open
  3443. \begin_layout Plain Layout
  3444. \align center
  3445. \begin_inset Graphics
  3446. filename graphics/Globin Paper/figure3 - detection.pdf
  3447. \end_inset
  3448. \end_layout
  3449. \begin_layout Plain Layout
  3450. \begin_inset Caption Standard
  3451. \begin_layout Plain Layout
  3452. \series bold
  3453. \begin_inset Argument 1
  3454. status collapsed
  3455. \begin_layout Plain Layout
  3456. Gene detections as a function of abundance thresholds in globin-blocked
  3457. (GB) and non-GB samples.
  3458. \end_layout
  3459. \end_inset
  3460. \begin_inset CommandInset label
  3461. LatexCommand label
  3462. name "fig:Gene-detections"
  3463. \end_inset
  3464. Gene detections as a function of abundance thresholds in globin-blocked
  3465. (GB) and non-GB samples.
  3466. \series default
  3467. Average abundance (logCPM,
  3468. \begin_inset Formula $\log_{2}$
  3469. \end_inset
  3470. counts per million reads counted) was computed by separate group normalization
  3471. as described in Figure
  3472. \begin_inset CommandInset ref
  3473. LatexCommand ref
  3474. reference "fig:logcpm-dists"
  3475. plural "false"
  3476. caps "false"
  3477. noprefix "false"
  3478. \end_inset
  3479. for both the GB and non-GB groups, as well as for all samples considered
  3480. as one large group.
  3481. For each every integer threshold from -2 to 3, the number of genes detected
  3482. at or above that logCPM threshold was plotted for each group.
  3483. \end_layout
  3484. \end_inset
  3485. \end_layout
  3486. \begin_layout Plain Layout
  3487. \end_layout
  3488. \end_inset
  3489. \end_layout
  3490. \begin_layout Standard
  3491. Based on these distributions, we selected a detection threshold of -1, which
  3492. is approximately the leftmost edge of the trough between the signal and
  3493. noise peaks.
  3494. This represents the most liberal possible detection threshold that doesn't
  3495. call substantial numbers of noise genes as detected.
  3496. Among the full dataset, 13429 genes were detected at this threshold, and
  3497. 22276 were not.
  3498. When considering the GB libraries and non-GB libraries separately and re-comput
  3499. ing normalization factors independently within each group, 14535 genes were
  3500. detected in the GB libraries while only 12460 were detected in the non-GB
  3501. libraries.
  3502. Thus, GB allowed the detection of 2000 extra genes that were buried under
  3503. the noise floor without GB.
  3504. This pattern of at least 2000 additional genes detected with GB was also
  3505. consistent across a wide range of possible detection thresholds, from -2
  3506. to 3 (see Figure
  3507. \begin_inset CommandInset ref
  3508. LatexCommand ref
  3509. reference "fig:Gene-detections"
  3510. plural "false"
  3511. caps "false"
  3512. noprefix "false"
  3513. \end_inset
  3514. ).
  3515. \end_layout
  3516. \begin_layout Subsection*
  3517. Globin blocking does not add significant additional noise or decrease sample
  3518. quality
  3519. \end_layout
  3520. \begin_layout Standard
  3521. One potential worry is that the globin blocking protocol could perturb the
  3522. levels of non-globin genes.
  3523. There are two kinds of possible perturbations: systematic and random.
  3524. The former is not a major concern for detection of differential expression,
  3525. since a 2-fold change in every sample has no effect on the relative fold
  3526. change between samples.
  3527. In contrast, random perturbations would increase the noise and obscure
  3528. the signal in the dataset, reducing the capacity to detect differential
  3529. expression.
  3530. \end_layout
  3531. \begin_layout Standard
  3532. \begin_inset Float figure
  3533. wide false
  3534. sideways false
  3535. status open
  3536. \begin_layout Plain Layout
  3537. \align center
  3538. \begin_inset Graphics
  3539. filename graphics/Globin Paper/figure4 - maplot-colored.pdf
  3540. \end_inset
  3541. \end_layout
  3542. \begin_layout Plain Layout
  3543. \begin_inset Caption Standard
  3544. \begin_layout Plain Layout
  3545. \begin_inset Argument 1
  3546. status collapsed
  3547. \begin_layout Plain Layout
  3548. MA plot showing effects of globin blocking on each gene's abundance.
  3549. \end_layout
  3550. \end_inset
  3551. \begin_inset CommandInset label
  3552. LatexCommand label
  3553. name "fig:MA-plot"
  3554. \end_inset
  3555. \series bold
  3556. MA plot showing effects of globin blocking on each gene's abundance.
  3557. \series default
  3558. All libraries were normalized together as described in Figure
  3559. \begin_inset CommandInset ref
  3560. LatexCommand ref
  3561. reference "fig:logcpm-dists"
  3562. plural "false"
  3563. caps "false"
  3564. noprefix "false"
  3565. \end_inset
  3566. , and genes with an average logCPM below -1 were filtered out.
  3567. Each remaining gene was tested for differential abundance with respect
  3568. to globin blocking (GB) using edgeR’s quasi-likelihod F-test, fitting a
  3569. negative binomial generalized linear model to table of read counts in each
  3570. library.
  3571. For each gene, edgeR reported average abundance (logCPM),
  3572. \begin_inset Formula $\log_{2}$
  3573. \end_inset
  3574. fold change (logFC), p-value, and Benjamini-Hochberg adjusted false discovery
  3575. rate (FDR).
  3576. Each gene's logFC was plotted against its logCPM, colored by FDR.
  3577. Red points are significant at ≤10% FDR, and blue are not significant at
  3578. that threshold.
  3579. The alpha and beta globin genes targeted for blocking are marked with large
  3580. triangles, while all other genes are represented as small points.
  3581. \end_layout
  3582. \end_inset
  3583. \end_layout
  3584. \begin_layout Plain Layout
  3585. \end_layout
  3586. \end_inset
  3587. \end_layout
  3588. \begin_layout Standard
  3589. \begin_inset Flex TODO Note (inline)
  3590. status open
  3591. \begin_layout Plain Layout
  3592. Standardize on
  3593. \begin_inset Quotes eld
  3594. \end_inset
  3595. log2
  3596. \begin_inset Quotes erd
  3597. \end_inset
  3598. notation
  3599. \end_layout
  3600. \end_inset
  3601. \end_layout
  3602. \begin_layout Standard
  3603. The data do indeed show small systematic perturbations in gene levels (Figure
  3604. \begin_inset CommandInset ref
  3605. LatexCommand ref
  3606. reference "fig:MA-plot"
  3607. plural "false"
  3608. caps "false"
  3609. noprefix "false"
  3610. \end_inset
  3611. ).
  3612. Other than the 3 designated alpha and beta globin genes, two other genes
  3613. stand out as having especially large negative log fold changes: HBD and
  3614. LOC1021365.
  3615. HBD, delta globin, is most likely targeted by the blocking oligos due to
  3616. high sequence homology with the other globin genes.
  3617. LOC1021365 is the aforementioned ncRNA that is reverse-complementary to
  3618. one of the alpha-like genes and that would be expected to be removed during
  3619. the globin blocking step.
  3620. All other genes appear in a cluster centered vertically at 0, and the vast
  3621. majority of genes in this cluster show an absolute log2(FC) of 0.5 or less.
  3622. Nevertheless, many of these small perturbations are still statistically
  3623. significant, indicating that the globin blocking oligos likely cause very
  3624. small but non-zero systematic perturbations in measured gene expression
  3625. levels.
  3626. \end_layout
  3627. \begin_layout Standard
  3628. \begin_inset Float figure
  3629. wide false
  3630. sideways false
  3631. status open
  3632. \begin_layout Plain Layout
  3633. \align center
  3634. \begin_inset Graphics
  3635. filename graphics/Globin Paper/figure5 - corrplot.pdf
  3636. \end_inset
  3637. \end_layout
  3638. \begin_layout Plain Layout
  3639. \begin_inset Caption Standard
  3640. \begin_layout Plain Layout
  3641. \series bold
  3642. \begin_inset Argument 1
  3643. status collapsed
  3644. \begin_layout Plain Layout
  3645. Comparison of inter-sample gene abundance correlations with and without
  3646. globin blocking.
  3647. \end_layout
  3648. \end_inset
  3649. \begin_inset CommandInset label
  3650. LatexCommand label
  3651. name "fig:gene-abundance-correlations"
  3652. \end_inset
  3653. Comparison of inter-sample gene abundance correlations with and without
  3654. globin blocking (GB).
  3655. \series default
  3656. All libraries were normalized together as described in Figure 2, and genes
  3657. with an average abundance (logCPM, log2 counts per million reads counted)
  3658. less than -1 were filtered out.
  3659. Each gene’s logCPM was computed in each library using the edgeR cpm function.
  3660. For each pair of biological samples, the Pearson correlation between those
  3661. samples' GB libraries was plotted against the correlation between the same
  3662. samples’ non-GB libraries.
  3663. Each point represents an unique pair of samples.
  3664. The solid gray line shows a quantile-quantile plot of distribution of GB
  3665. correlations vs.
  3666. that of non-GB correlations.
  3667. The thin dashed line is the identity line, provided for reference.
  3668. \end_layout
  3669. \end_inset
  3670. \end_layout
  3671. \begin_layout Plain Layout
  3672. \end_layout
  3673. \end_inset
  3674. \end_layout
  3675. \begin_layout Standard
  3676. To evaluate the possibility of globin blocking causing random perturbations
  3677. and reducing sample quality, we computed the Pearson correlation between
  3678. logCPM values for every pair of samples with and without GB and plotted
  3679. them against each other (Figure
  3680. \begin_inset CommandInset ref
  3681. LatexCommand ref
  3682. reference "fig:gene-abundance-correlations"
  3683. plural "false"
  3684. caps "false"
  3685. noprefix "false"
  3686. \end_inset
  3687. ).
  3688. The plot indicated that the GB libraries have higher sample-to-sample correlati
  3689. ons than the non-GB libraries.
  3690. Parametric and nonparametric tests for differences between the correlations
  3691. with and without GB both confirmed that this difference was highly significant
  3692. (2-sided paired t-test: t = 37.2, df = 665, P ≪ 2.2e-16; 2-sided Wilcoxon
  3693. sign-rank test: V = 2195, P ≪ 2.2e-16).
  3694. Performing the same tests on the Spearman correlations gave the same conclusion
  3695. (t-test: t = 26.8, df = 665, P ≪ 2.2e-16; sign-rank test: V = 8781, P ≪ 2.2e-16).
  3696. The edgeR package was used to compute the overall biological coefficient
  3697. of variation (BCV) for GB and non-GB libraries, and found that globin blocking
  3698. resulted in a negligible increase in the BCV (0.417 with GB vs.
  3699. 0.400 without).
  3700. The near equality of the BCVs for both sets indicates that the higher correlati
  3701. ons in the GB libraries are most likely a result of the increased yield
  3702. of useful reads, which reduces the contribution of Poisson counting uncertainty
  3703. to the overall variance of the logCPM values
  3704. \begin_inset CommandInset citation
  3705. LatexCommand cite
  3706. key "McCarthy2012"
  3707. literal "false"
  3708. \end_inset
  3709. .
  3710. This improves the precision of expression measurements and more than offsets
  3711. the negligible increase in BCV.
  3712. \end_layout
  3713. \begin_layout Subsection*
  3714. More differentially expressed genes are detected with globin blocking
  3715. \end_layout
  3716. \begin_layout Standard
  3717. \begin_inset Float table
  3718. wide false
  3719. sideways false
  3720. status open
  3721. \begin_layout Plain Layout
  3722. \align center
  3723. \begin_inset Tabular
  3724. <lyxtabular version="3" rows="5" columns="5">
  3725. <features tabularvalignment="middle">
  3726. <column alignment="center" valignment="top">
  3727. <column alignment="center" valignment="top">
  3728. <column alignment="center" valignment="top">
  3729. <column alignment="center" valignment="top">
  3730. <column alignment="center" valignment="top">
  3731. <row>
  3732. <cell alignment="center" valignment="top" usebox="none">
  3733. \begin_inset Text
  3734. \begin_layout Plain Layout
  3735. \end_layout
  3736. \end_inset
  3737. </cell>
  3738. <cell alignment="center" valignment="top" usebox="none">
  3739. \begin_inset Text
  3740. \begin_layout Plain Layout
  3741. \end_layout
  3742. \end_inset
  3743. </cell>
  3744. <cell multicolumn="1" alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3745. \begin_inset Text
  3746. \begin_layout Plain Layout
  3747. \series bold
  3748. No Globin Blocking
  3749. \end_layout
  3750. \end_inset
  3751. </cell>
  3752. <cell multicolumn="2" alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3753. \begin_inset Text
  3754. \begin_layout Plain Layout
  3755. \end_layout
  3756. \end_inset
  3757. </cell>
  3758. <cell multicolumn="2" alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  3759. \begin_inset Text
  3760. \begin_layout Plain Layout
  3761. \end_layout
  3762. \end_inset
  3763. </cell>
  3764. </row>
  3765. <row>
  3766. <cell alignment="center" valignment="top" usebox="none">
  3767. \begin_inset Text
  3768. \begin_layout Plain Layout
  3769. \end_layout
  3770. \end_inset
  3771. </cell>
  3772. <cell alignment="center" valignment="top" usebox="none">
  3773. \begin_inset Text
  3774. \begin_layout Plain Layout
  3775. \end_layout
  3776. \end_inset
  3777. </cell>
  3778. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3779. \begin_inset Text
  3780. \begin_layout Plain Layout
  3781. \series bold
  3782. Up
  3783. \end_layout
  3784. \end_inset
  3785. </cell>
  3786. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3787. \begin_inset Text
  3788. \begin_layout Plain Layout
  3789. \series bold
  3790. NS
  3791. \end_layout
  3792. \end_inset
  3793. </cell>
  3794. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3795. \begin_inset Text
  3796. \begin_layout Plain Layout
  3797. \series bold
  3798. Down
  3799. \end_layout
  3800. \end_inset
  3801. </cell>
  3802. </row>
  3803. <row>
  3804. <cell multirow="3" alignment="center" valignment="middle" topline="true" bottomline="true" leftline="true" usebox="none">
  3805. \begin_inset Text
  3806. \begin_layout Plain Layout
  3807. \series bold
  3808. Globin-Blocking
  3809. \end_layout
  3810. \end_inset
  3811. </cell>
  3812. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3813. \begin_inset Text
  3814. \begin_layout Plain Layout
  3815. \series bold
  3816. Up
  3817. \end_layout
  3818. \end_inset
  3819. </cell>
  3820. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3821. \begin_inset Text
  3822. \begin_layout Plain Layout
  3823. \family roman
  3824. \series medium
  3825. \shape up
  3826. \size normal
  3827. \emph off
  3828. \bar no
  3829. \strikeout off
  3830. \xout off
  3831. \uuline off
  3832. \uwave off
  3833. \noun off
  3834. \color none
  3835. 231
  3836. \end_layout
  3837. \end_inset
  3838. </cell>
  3839. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3840. \begin_inset Text
  3841. \begin_layout Plain Layout
  3842. \family roman
  3843. \series medium
  3844. \shape up
  3845. \size normal
  3846. \emph off
  3847. \bar no
  3848. \strikeout off
  3849. \xout off
  3850. \uuline off
  3851. \uwave off
  3852. \noun off
  3853. \color none
  3854. 515
  3855. \end_layout
  3856. \end_inset
  3857. </cell>
  3858. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3859. \begin_inset Text
  3860. \begin_layout Plain Layout
  3861. \family roman
  3862. \series medium
  3863. \shape up
  3864. \size normal
  3865. \emph off
  3866. \bar no
  3867. \strikeout off
  3868. \xout off
  3869. \uuline off
  3870. \uwave off
  3871. \noun off
  3872. \color none
  3873. 2
  3874. \end_layout
  3875. \end_inset
  3876. </cell>
  3877. </row>
  3878. <row>
  3879. <cell multirow="4" alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3880. \begin_inset Text
  3881. \begin_layout Plain Layout
  3882. \end_layout
  3883. \end_inset
  3884. </cell>
  3885. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3886. \begin_inset Text
  3887. \begin_layout Plain Layout
  3888. \series bold
  3889. NS
  3890. \end_layout
  3891. \end_inset
  3892. </cell>
  3893. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3894. \begin_inset Text
  3895. \begin_layout Plain Layout
  3896. \family roman
  3897. \series medium
  3898. \shape up
  3899. \size normal
  3900. \emph off
  3901. \bar no
  3902. \strikeout off
  3903. \xout off
  3904. \uuline off
  3905. \uwave off
  3906. \noun off
  3907. \color none
  3908. 160
  3909. \end_layout
  3910. \end_inset
  3911. </cell>
  3912. <cell alignment="center" valignment="top" topline="true" leftline="true" usebox="none">
  3913. \begin_inset Text
  3914. \begin_layout Plain Layout
  3915. \family roman
  3916. \series medium
  3917. \shape up
  3918. \size normal
  3919. \emph off
  3920. \bar no
  3921. \strikeout off
  3922. \xout off
  3923. \uuline off
  3924. \uwave off
  3925. \noun off
  3926. \color none
  3927. 11235
  3928. \end_layout
  3929. \end_inset
  3930. </cell>
  3931. <cell alignment="center" valignment="top" topline="true" leftline="true" rightline="true" usebox="none">
  3932. \begin_inset Text
  3933. \begin_layout Plain Layout
  3934. \family roman
  3935. \series medium
  3936. \shape up
  3937. \size normal
  3938. \emph off
  3939. \bar no
  3940. \strikeout off
  3941. \xout off
  3942. \uuline off
  3943. \uwave off
  3944. \noun off
  3945. \color none
  3946. 136
  3947. \end_layout
  3948. \end_inset
  3949. </cell>
  3950. </row>
  3951. <row>
  3952. <cell multirow="4" alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3953. \begin_inset Text
  3954. \begin_layout Plain Layout
  3955. \end_layout
  3956. \end_inset
  3957. </cell>
  3958. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3959. \begin_inset Text
  3960. \begin_layout Plain Layout
  3961. \series bold
  3962. Down
  3963. \end_layout
  3964. \end_inset
  3965. </cell>
  3966. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3967. \begin_inset Text
  3968. \begin_layout Plain Layout
  3969. \family roman
  3970. \series medium
  3971. \shape up
  3972. \size normal
  3973. \emph off
  3974. \bar no
  3975. \strikeout off
  3976. \xout off
  3977. \uuline off
  3978. \uwave off
  3979. \noun off
  3980. \color none
  3981. 0
  3982. \end_layout
  3983. \end_inset
  3984. </cell>
  3985. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" usebox="none">
  3986. \begin_inset Text
  3987. \begin_layout Plain Layout
  3988. \family roman
  3989. \series medium
  3990. \shape up
  3991. \size normal
  3992. \emph off
  3993. \bar no
  3994. \strikeout off
  3995. \xout off
  3996. \uuline off
  3997. \uwave off
  3998. \noun off
  3999. \color none
  4000. 548
  4001. \end_layout
  4002. \end_inset
  4003. </cell>
  4004. <cell alignment="center" valignment="top" topline="true" bottomline="true" leftline="true" rightline="true" usebox="none">
  4005. \begin_inset Text
  4006. \begin_layout Plain Layout
  4007. \family roman
  4008. \series medium
  4009. \shape up
  4010. \size normal
  4011. \emph off
  4012. \bar no
  4013. \strikeout off
  4014. \xout off
  4015. \uuline off
  4016. \uwave off
  4017. \noun off
  4018. \color none
  4019. 127
  4020. \end_layout
  4021. \end_inset
  4022. </cell>
  4023. </row>
  4024. </lyxtabular>
  4025. \end_inset
  4026. \end_layout
  4027. \begin_layout Plain Layout
  4028. \begin_inset Caption Standard
  4029. \begin_layout Plain Layout
  4030. \series bold
  4031. \begin_inset Argument 1
  4032. status open
  4033. \begin_layout Plain Layout
  4034. Comparison of significantly differentially expressed genes with and without
  4035. globin blocking.
  4036. \end_layout
  4037. \end_inset
  4038. \begin_inset CommandInset label
  4039. LatexCommand label
  4040. name "tab:Comparison-of-significant"
  4041. \end_inset
  4042. Comparison of significantly differentially expressed genes with and without
  4043. globin blocking.
  4044. \series default
  4045. Up, Down: Genes significantly up/down-regulated in post-transplant samples
  4046. relative to pre-transplant samples, with a false discovery rate of 10%
  4047. or less.
  4048. NS: Non-significant genes (false discovery rate greater than 10%).
  4049. \end_layout
  4050. \end_inset
  4051. \end_layout
  4052. \begin_layout Plain Layout
  4053. \end_layout
  4054. \end_inset
  4055. \end_layout
  4056. \begin_layout Standard
  4057. To compare performance on differential gene expression tests, we took subsets
  4058. of both the GB and non-GB libraries with exactly one pre-transplant and
  4059. one post-transplant sample for each animal that had paired samples available
  4060. for analysis (N=7 animals, N=14 samples in each subset).
  4061. The same test for pre- vs.
  4062. post-transplant differential gene expression was performed on the same
  4063. 7 pairs of samples from GB libraries and non-GB libraries, in each case
  4064. using an FDR of 10% as the threshold of significance.
  4065. Out of 12954 genes that passed the detection threshold in both subsets,
  4066. 358 were called significantly differentially expressed in the same direction
  4067. in both sets; 1063 were differentially expressed in the GB set only; 296
  4068. were differentially expressed in the non-GB set only; 2 genes were called
  4069. significantly up in the GB set but significantly down in the non-GB set;
  4070. and the remaining 11235 were not called differentially expressed in either
  4071. set.
  4072. These data are summarized in Table
  4073. \begin_inset CommandInset ref
  4074. LatexCommand ref
  4075. reference "tab:Comparison-of-significant"
  4076. plural "false"
  4077. caps "false"
  4078. noprefix "false"
  4079. \end_inset
  4080. .
  4081. The differences in BCV calculated by EdgeR for these subsets of samples
  4082. were negligible (BCV = 0.302 for GB and 0.297 for non-GB).
  4083. \end_layout
  4084. \begin_layout Standard
  4085. The key point is that the GB data results in substantially more differentially
  4086. expressed calls than the non-GB data.
  4087. Since there is no gold standard for this dataset, it is impossible to be
  4088. certain whether this is due to under-calling of differential expression
  4089. in the non-GB samples or over-calling in the GB samples.
  4090. However, given that both datasets are derived from the same biological
  4091. samples and have nearly equal BCVs, it is more likely that the larger number
  4092. of DE calls in the GB samples are genuine detections that were enabled
  4093. by the higher sequencing depth and measurement precision of the GB samples.
  4094. Note that the same set of genes was considered in both subsets, so the
  4095. larger number of differentially expressed gene calls in the GB data set
  4096. reflects a greater sensitivity to detect significant differential gene
  4097. expression and not simply the larger total number of detected genes in
  4098. GB samples described earlier.
  4099. \end_layout
  4100. \begin_layout Section
  4101. Discussion
  4102. \end_layout
  4103. \begin_layout Standard
  4104. The original experience with whole blood gene expression profiling on DNA
  4105. microarrays demonstrated that the high concentration of globin transcripts
  4106. reduced the sensitivity to detect genes with relatively low expression
  4107. levels, in effect, significantly reducing the sensitivity.
  4108. To address this limitation, commercial protocols for globin reduction were
  4109. developed based on strategies to block globin transcript amplification
  4110. during labeling or physically removing globin transcripts by affinity bead
  4111. methods
  4112. \begin_inset CommandInset citation
  4113. LatexCommand cite
  4114. key "Winn2010"
  4115. literal "false"
  4116. \end_inset
  4117. .
  4118. More recently, using the latest generation of labeling protocols and arrays,
  4119. it was determined that globin reduction was no longer necessary to obtain
  4120. sufficient sensitivity to detect differential transcript expression
  4121. \begin_inset CommandInset citation
  4122. LatexCommand cite
  4123. key "NuGEN2010"
  4124. literal "false"
  4125. \end_inset
  4126. .
  4127. However, we are not aware of any publications using these currently available
  4128. protocols the with latest generation of microarrays that actually compare
  4129. the detection sensitivity with and without globin reduction.
  4130. However, in practice this has now been adopted generally primarily driven
  4131. by concerns for cost control.
  4132. The main objective of our work was to directly test the impact of globin
  4133. gene transcripts and a new globin blocking protocol for application to
  4134. the newest generation of differential gene expression profiling determined
  4135. using next generation sequencing.
  4136. \end_layout
  4137. \begin_layout Standard
  4138. The challenge of doing global gene expression profiling in cynomolgus monkeys
  4139. is that the current available arrays were never designed to comprehensively
  4140. cover this genome and have not been updated since the first assemblies
  4141. of the cynomolgus genome were published.
  4142. Therefore, we determined that the best strategy for peripheral blood profiling
  4143. was to do deep RNA-seq and inform the workflow using the latest available
  4144. genome assembly and annotation
  4145. \begin_inset CommandInset citation
  4146. LatexCommand cite
  4147. key "Wilson2013"
  4148. literal "false"
  4149. \end_inset
  4150. .
  4151. However, it was not immediately clear whether globin reduction was necessary
  4152. for RNA-seq or how much improvement in efficiency or sensitivity to detect
  4153. differential gene expression would be achieved for the added cost and work.
  4154. \end_layout
  4155. \begin_layout Standard
  4156. We only found one report that demonstrated that globin reduction significantly
  4157. improved the effective read yields for sequencing of human peripheral blood
  4158. cell RNA using a DeepSAGE protocol
  4159. \begin_inset CommandInset citation
  4160. LatexCommand cite
  4161. key "Mastrokolias2012"
  4162. literal "false"
  4163. \end_inset
  4164. .
  4165. The approach to DeepSAGE involves two different restriction enzymes that
  4166. purify and then tag small fragments of transcripts at specific locations
  4167. and thus, significantly reduces the complexity of the transcriptome.
  4168. Therefore, we could not determine how DeepSAGE results would translate
  4169. to the common strategy in the field for assaying the entire transcript
  4170. population by whole-transcriptome 3’-end RNA-seq.
  4171. Furthermore, if globin reduction is necessary, we also needed a globin
  4172. reduction method specific to cynomolgus globin sequences that would work
  4173. an organism for which no kit is available off the shelf.
  4174. \end_layout
  4175. \begin_layout Standard
  4176. As mentioned above, the addition of globin blocking oligos has a very small
  4177. impact on measured expression levels of gene expression.
  4178. However, this is a non-issue for the purposes of differential expression
  4179. testing, since a systematic change in a gene in all samples does not affect
  4180. relative expression levels between samples.
  4181. However, we must acknowledge that simple comparisons of gene expression
  4182. data obtained by GB and non-GB protocols are not possible without additional
  4183. normalization.
  4184. \end_layout
  4185. \begin_layout Standard
  4186. More importantly, globin blocking not only nearly doubles the yield of usable
  4187. reads, it also increases inter-sample correlation and sensitivity to detect
  4188. differential gene expression relative to the same set of samples profiled
  4189. without blocking.
  4190. In addition, globin blocking does not add a significant amount of random
  4191. noise to the data.
  4192. Globin blocking thus represents a cost-effective way to squeeze more data
  4193. and statistical power out of the same blood samples and the same amount
  4194. of sequencing.
  4195. In conclusion, globin reduction greatly increases the yield of useful RNA-seq
  4196. reads mapping to the rest of the genome, with minimal perturbations in
  4197. the relative levels of non-globin genes.
  4198. Based on these results, globin transcript reduction using sequence-specific,
  4199. complementary blocking oligonucleotides is recommended for all deep RNA-seq
  4200. of cynomolgus and other nonhuman primate blood samples.
  4201. \end_layout
  4202. \begin_layout Chapter
  4203. Future Directions
  4204. \end_layout
  4205. \begin_layout Itemize
  4206. Study other epigenetic marks in more contexts
  4207. \end_layout
  4208. \begin_deeper
  4209. \begin_layout Itemize
  4210. DNA methylation, histone marks, chromatin accessibility & conformation in
  4211. CD4 T-cells
  4212. \end_layout
  4213. \begin_layout Itemize
  4214. Also look at other types lymphocytes: CD8 T-cells, B-cells, NK cells
  4215. \end_layout
  4216. \end_deeper
  4217. \begin_layout Itemize
  4218. Use CV or bootstrap to better evaluate classifiers
  4219. \end_layout
  4220. \begin_layout Standard
  4221. \begin_inset ERT
  4222. status open
  4223. \begin_layout Plain Layout
  4224. % Use "References" instead of "Bibliography"
  4225. \end_layout
  4226. \begin_layout Plain Layout
  4227. \backslash
  4228. renewcommand{
  4229. \backslash
  4230. bibname}{References}
  4231. \end_layout
  4232. \end_inset
  4233. \end_layout
  4234. \begin_layout Standard
  4235. \begin_inset Flex TODO Note (inline)
  4236. status open
  4237. \begin_layout Plain Layout
  4238. Check bib entry formatting & sort order
  4239. \end_layout
  4240. \end_inset
  4241. \end_layout
  4242. \begin_layout Standard
  4243. \begin_inset CommandInset bibtex
  4244. LatexCommand bibtex
  4245. btprint "btPrintCited"
  4246. bibfiles "refs"
  4247. options "bibtotoc,unsrt"
  4248. \end_inset
  4249. \end_layout
  4250. \end_body
  4251. \end_document