#LyX 2.3 created this file. For more info see http://www.lyx.org/ \lyxformat 544 \begin_document \begin_header \save_transient_properties true \origin unavailable \textclass extbook \begin_preamble % List all used files in log output \listfiles % Add a DRAFT watermark \usepackage{draftwatermark} \SetWatermarkLightness{0.97} \SetWatermarkScale{1} % Set up required header format \usepackage{fancyhdr} \pagestyle{fancy} \renewcommand{\headrulewidth}{0pt} \rhead{} \lhead{} \rfoot{} \lfoot{} \cfoot{\thepage} % Page number bottom center % https://tex.stackexchange.com/questions/65680/automatically-bold-first-sentence-of-a-floats-caption \usepackage{xstring} \usepackage{etoolbox} \usepackage{caption} \captionsetup{labelfont=bf,tableposition=top} \makeatletter \newcommand\formatlabel[1]{% \noexpandarg \IfSubStr{#1}{.}{% \StrBefore{#1}{.}[\firstcaption]% \StrBehind{#1}{.}[\secondcaption]% \textbf{\firstcaption.} \secondcaption}{% #1}% } \patchcmd{\@caption}{#3}{\formatlabel{#3}} \makeatother \end_preamble \use_default_options true \begin_modules todonotes \end_modules \maintain_unincluded_children false \language english \language_package default \inputencoding utf8 \fontencoding default \font_roman "default" "default" \font_sans "default" "default" \font_typewriter "default" "default" \font_math "auto" "auto" \font_default_family default \use_non_tex_fonts false \font_sc false \font_osf false \font_sf_scale 100 100 \font_tt_scale 100 100 \use_microtype false \use_dash_ligatures true \graphics default \default_output_format pdf4 \output_sync 0 \bibtex_command default \index_command default \paperfontsize 12 \spacing double \use_hyperref true \pdf_bookmarks true \pdf_bookmarksnumbered false \pdf_bookmarksopen false \pdf_bookmarksopenlevel 1 \pdf_breaklinks false \pdf_pdfborder false \pdf_colorlinks false \pdf_backref false \pdf_pdfusetitle true \papersize letterpaper \use_geometry true \use_package amsmath 1 \use_package amssymb 1 \use_package cancel 1 \use_package esint 1 \use_package mathdots 1 \use_package mathtools 1 \use_package mhchem 1 \use_package stackrel 1 \use_package stmaryrd 1 \use_package undertilde 1 \cite_engine basic \cite_engine_type default \biblio_style plain \use_bibtopic false \use_indices false \paperorientation portrait \suppress_date false \justification true \use_refstyle 1 \use_minted 0 \index Index \shortcut idx \color #008000 \end_index \leftmargin 1.5in \topmargin 1in \rightmargin 1in \bottommargin 1in \secnumdepth 3 \tocdepth 3 \paragraph_separation indent \paragraph_indentation default \is_math_indent 0 \math_numbering_side default \quotes_style english \dynamic_quotes 0 \papercolumns 1 \papersides 2 \paperpagestyle default \tracking_changes false \output_changes false \html_math_output 0 \html_css_as_file 0 \html_be_strict false \end_header \begin_body \begin_layout Title Bioinformatic analysis of complex, high-throughput genomic and epigenomic data in the context of immunology and transplant rejection \end_layout \begin_layout Author A thesis presented \begin_inset Newline newline \end_inset by \begin_inset Newline newline \end_inset Ryan C. Thompson \begin_inset Newline newline \end_inset to \begin_inset Newline newline \end_inset The Scripps Research Institute Graduate Program \begin_inset Newline newline \end_inset in partial fulfillment of the requirements for the degree of \begin_inset Newline newline \end_inset Doctor of Philosophy in the subject of Biology \begin_inset Newline newline \end_inset for \begin_inset Newline newline \end_inset The Scripps Research Institute \begin_inset Newline newline \end_inset La Jolla, California \end_layout \begin_layout Date May 2019 \end_layout \begin_layout Standard [Copyright notice] \end_layout \begin_layout Standard [Thesis acceptance form] \end_layout \begin_layout Standard [Dedication] \end_layout \begin_layout Standard [Acknowledgements] \end_layout \begin_layout Standard \begin_inset CommandInset toc LatexCommand tableofcontents \end_inset \end_layout \begin_layout Standard \begin_inset FloatList table \end_inset \end_layout \begin_layout Standard \begin_inset FloatList figure \end_inset \end_layout \begin_layout Standard [List of Abbreviations] \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Look into auto-generated nomenclature list: https://wiki.lyx.org/Tips/Nomenclature \end_layout \end_inset \end_layout \begin_layout List of TODOs \end_layout \begin_layout Standard [Abstract] \end_layout \begin_layout Chapter* Abstract \end_layout \begin_layout Chapter Introduction \end_layout \begin_layout Section Background & Significance \end_layout \begin_layout Subsection Biological motivation \end_layout \begin_layout Itemize Rejection is the major long-term threat to organ and tissue grafts \end_layout \begin_deeper \begin_layout Itemize Common mechanisms of rejection \end_layout \begin_layout Itemize Effective immune suppression requires monitoring for rejection and tuning \end_layout \begin_layout Itemize Current tests for rejection (tissue biopsy) are invasive and biased \end_layout \begin_layout Itemize A blood test based on microarrays would be less biased and invasive \end_layout \end_deeper \begin_layout Itemize Memory cells are resistant to immune suppression \end_layout \begin_deeper \begin_layout Itemize Mechanisms of resistance in memory cells are poorly understood \end_layout \begin_layout Itemize A better understanding of immune memory formation is needed \end_layout \end_deeper \begin_layout Itemize Mesenchymal stem cell infusion is a promising new treatment to prevent/delay rejection \end_layout \begin_deeper \begin_layout Itemize Demonstrated in mice, but not yet in primates \end_layout \begin_layout Itemize Mechanism currently unknown, but MSC are known to be immune modulatory \end_layout \end_deeper \begin_layout Subsection Overview of bioinformatic analysis methods \end_layout \begin_layout Standard An overview of all the methods used, including what problem they solve, what assumptions they make, and a basic description of how they work. \end_layout \begin_layout Itemize ChIP-seq Peak calling \end_layout \begin_deeper \begin_layout Itemize Cross-correlation analysis to determine fragment size \end_layout \begin_layout Itemize Broad vs narrow peaks \end_layout \begin_layout Itemize SICER for broad peaks \end_layout \begin_layout Itemize IDR for biologically reproducible peaks \end_layout \begin_layout Itemize csaw peak filtering guidelines for unbiased downstream analysis \end_layout \end_deeper \begin_layout Itemize Normalization is non-trivial and application-dependant \end_layout \begin_deeper \begin_layout Itemize Expression arrays: RMA & fRMA; why fRMA is needed \end_layout \begin_layout Itemize Methylation arrays: M-value transformation approximates normal data but induces heteroskedasticity \end_layout \begin_layout Itemize RNA-seq: normalize based on assumption that the average gene is not changing \end_layout \begin_layout Itemize ChIP-seq: complex with many considerations, dependent on experimental methods, biological system, and analysis goals \end_layout \end_deeper \begin_layout Itemize Limma: The standard linear modeling framework for genomics \end_layout \begin_deeper \begin_layout Itemize empirical Bayes variance modeling: limma's core feature \end_layout \begin_layout Itemize edgeR & DESeq2: Extend with negative bonomial GLM for RNA-seq and other count data \end_layout \begin_layout Itemize voom: Extend with precision weights to model mean-variance trend \end_layout \begin_layout Itemize arrayWeights and duplicateCorrelation to handle complex variance structures \end_layout \end_deeper \begin_layout Itemize sva and ComBat for batch correction \end_layout \begin_layout Itemize Factor analysis: PCA, MDS, MOFA \end_layout \begin_deeper \begin_layout Itemize Batch-corrected PCA is informative, but careful application is required to avoid bias \end_layout \end_deeper \begin_layout Itemize Gene set analysis: camera and SPIA \end_layout \begin_layout Section Innovation \end_layout \begin_layout Itemize MSC infusion to improve transplant outcomes (prevent/delay rejection) \end_layout \begin_deeper \begin_layout Itemize Characterize MSC response to interferon gamma \end_layout \begin_layout Itemize IFN-g is thought to stimulate their function \end_layout \begin_layout Itemize Test IFN-g treated MSC infusion as a therapy to delay graft rejection in cynomolgus monkeys \end_layout \begin_layout Itemize Monitor animals post-transplant using blood RNA-seq at serial time points \end_layout \end_deeper \begin_layout Itemize Investigate dynamics of histone marks in CD4 T-cell activation and memory \end_layout \begin_deeper \begin_layout Itemize Previous studies have looked at single snapshots of histone marks \end_layout \begin_layout Itemize Instead, look at changes in histone marks across activation and memory \end_layout \end_deeper \begin_layout Itemize High-throughput sequencing and microarray technologies \end_layout \begin_deeper \begin_layout Itemize Powerful methods for assaying gene expression and epigenetics across entire genomes \end_layout \begin_layout Itemize Proper analysis requires finding and exploiting systematic genome-wide trends \end_layout \end_deeper \begin_layout Chapter Reproducible genome-wide epigenetic analysis of H3K4 and H3K27 methylation in naive and memory CD4 T-cell activation \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Author list: Me, Sarah, Dan \end_layout \end_inset \end_layout \begin_layout Section Approach \end_layout \begin_layout Itemize CD4 T-cells are central to all adaptive immune responses and memory \end_layout \begin_layout Itemize H3K4 and H3K27 methylation are major epigenetic regulators of gene expression \end_layout \begin_layout Itemize Canonically, H3K4 is activating and H3K27 is inhibitory, but the reality is complex \end_layout \begin_layout Itemize Looking at these marks during CD4 activation and memory should reveal new mechanistic details \end_layout \begin_layout Itemize Test \begin_inset Quotes eld \end_inset poised promoter \begin_inset Quotes erd \end_inset hypothesis in which H3K4 and H3K27 are both methylated \end_layout \begin_layout Itemize Expand scope of analysis beyond simple promoter counts \end_layout \begin_deeper \begin_layout Itemize Analyze peaks genome-wide, including in intergenic regions \end_layout \begin_layout Itemize Analysis of coverage distribution shape within promoters, e.g. upstream vs downstream coverage \end_layout \end_deeper \begin_layout Section Methods \end_layout \begin_layout Itemize Re-analyze previously published CD4 ChIP-seq & RNA-seq data \begin_inset CommandInset citation LatexCommand cite key "LaMere2016,Lamere2017" literal "true" \end_inset \end_layout \begin_deeper \begin_layout Itemize Completely reimplement analysis from scratch as a reproducible workflow \end_layout \begin_layout Itemize Use newly published methods & algorithms not available during the original analysis: SICER, csaw, MOFA, ComBat, sva, GREAT, and more \end_layout \end_deeper \begin_layout Itemize SICER, IDR, csaw, & GREAT to call ChIP-seq peaks genome-wide, perform differenti al abundance analysis, and relate those peaks to gene expression \end_layout \begin_layout Itemize Promoter counts in sliding windows around each gene's highest-expressed TSS to investigate coverage distribution within promoters \end_layout \begin_layout Section Results \end_layout \begin_layout Standard \begin_inset Note Note status open \begin_layout Plain Layout Focus on what hypotheses were tested, then select figures that show how those hypotheses were tested, even if the result is a negative. \end_layout \end_inset \end_layout \begin_layout Subsection H3K4 and H3K27 methylation occur in broad regions and are enriched near promoters \end_layout \begin_layout Itemize Figures comparing MACS (non-broad peak caller) to SICER/epic (broad peak caller) \end_layout \begin_deeper \begin_layout Itemize Compare peak sizes and number of called peaks \end_layout \begin_layout Itemize Show representative IDR consistency plots for both \end_layout \end_deeper \begin_layout Itemize IDR analysis shows that SICER-called peaks are much more reproducible between biological replicates \end_layout \begin_layout Itemize Each histone mark is enriched within a certain radius of gene TSS positions, but that radius is different for each mark (figure) \end_layout \begin_layout Subsection RNA-seq has a large confounding batch effect \end_layout \begin_layout Itemize RNA-seq batch effect can be partially corrected, but still induces uncorrectable biases in downstream analysis \end_layout \begin_deeper \begin_layout Itemize Figure showing MDS plot before & after ComBat \end_layout \begin_layout Itemize Figure relating sample weights to batches, cell types, time points, etc., showing that one batch is significantly worse quality \end_layout \begin_layout Itemize Figures showing p-value histograms for within-batch and cross-batch contrasts, showing that cross-batch contrasts have attenuated signal, as do comparisons within the bad batch \end_layout \end_deeper \begin_layout Subsection ChIP-seq must be corrected for hidden confounding factors \end_layout \begin_layout Itemize Figures showing pre- and post-SVA MDS plots for each histone mark \end_layout \begin_layout Itemize Figures showing BCV plots with and without SVA for each histone mark \end_layout \begin_layout Subsection H3K4 and H3K27 promoter methylation has broadly the expected correlation with gene expression \end_layout \begin_layout Itemize H3K4 is correlated with higher expression, and H3K27 is correlated with lower expression genome-wide \end_layout \begin_layout Itemize Figures showing these correlations: box/violin plots of expression distributions with every combination of peak presence/absence in promoter \end_layout \begin_layout Itemize Appropriate statistical tests showing significant differences in expected directions \end_layout \begin_layout Subsection MOFA recovers biologically relevant variation from blind analysis by correlating across datasets \end_layout \begin_layout Itemize MOFA \begin_inset CommandInset citation LatexCommand cite key "Argelaguet2018" literal "false" \end_inset successfully separates biologically relevant patterns of variation from technical confounding factors without knowing the sample labels, by finding latent factors that explain variation across multiple data sets. \end_layout \begin_deeper \begin_layout Itemize Figure: show percent-variance-explained plot from MOFA and PCA-like plots for the relevant latent factors \end_layout \begin_layout Itemize MOFA analysis also shows that batch effect correction can't get much better than it already is (Figure comparing blind MOFA batch correction to ComBat correction) \end_layout \end_deeper \begin_layout Subsection Naive-to-memory convergence observed in H3K4 and RNA-seq data, not in H3K27me3 \end_layout \begin_layout Itemize H3K4 and RNA-seq data show clear evidence of naive convergence with memory between days 1 and 5 (MDS plot figure, also compare with last figure from \begin_inset CommandInset citation LatexCommand cite key "LaMere2016" literal "false" \end_inset ) \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Get explicit permission from Sarah to include the figure \end_layout \end_inset \end_layout \begin_layout Itemize Table of numbers of genes different between N & M at each time point, showing dwindling differences at later time points, consistent with convergence \end_layout \begin_layout Itemize Similar figure for H3K27me3 showing lack of convergence \end_layout \begin_layout Subsection Effect of promoter coverage upstream vs downstream of TSS \end_layout \begin_layout Itemize H3K4me peaks seem to correlate with increased expression as long as they are anywhere near the TSS \end_layout \begin_layout Itemize H3K27me3 peaks can have different correlations to gene expression depending on their position relative to TSS (e.g. upstream vs downstream) Results consistent with \begin_inset CommandInset citation LatexCommand cite key "Young2011" literal "false" \end_inset \end_layout \begin_layout Section Discussion \end_layout \begin_layout Itemize "Promoter radius" is not constant and must be defined empirically for a given data set \end_layout \begin_layout Itemize MOFA shows great promise for accelerating discovery of major biological effects in multi-omics datasets \end_layout \begin_deeper \begin_layout Itemize MOFA was added to this analysis late and played primarily a confirmatory role, but it was able to confirm earlier conclusions with much less prior information (no sample labels) and much less analyst effort \end_layout \begin_layout Itemize MOFA confirmed that the already-implemented batch correction in the RNA-seq data was already performing as well as possible given the limitations of the data \end_layout \end_deeper \begin_layout Itemize Naive-to-memory convergence implies that naive cells are differentiating into memory cells, and that gene expression and H3K4 methylation are involved in this differentiation while H3K27me3 is less involved \end_layout \begin_layout Itemize H3K27me3, canonically regarded as a deactivating mark, seems to have a more complex \end_layout \begin_layout Itemize Discuss advantages of developing using a reproducible workflow \end_layout \begin_layout Chapter Improving array-based analyses of transplant rejection by optimizing data preprocessing \end_layout \begin_layout Standard \begin_inset Note Note status open \begin_layout Plain Layout Author list: Me, Sunil, Tom, Padma, Dan \end_layout \end_inset \end_layout \begin_layout Section Approach \end_layout \begin_layout Subsection Proper pre-processing is essential for array data \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout This section could probably use some citations \end_layout \end_inset \end_layout \begin_layout Standard Microarrays, bead ararys, and similar assays produce raw data in the form of fluorescence intensity measurements, with the each intensity measurement proportional to the abundance of some fluorescently-labelled target DNA or RNA sequence that base pairs to a specific probe sequence. However, these measurements for each probe are also affected my many technical confounding factors, such as the concentration of target material, strength of off-target binding, and the sensitivity of the imaging sensor. Some array designs also use multiple probe sequences for each target. Hence, extensive pre-processing of array data is necessary to normalize out the effects of these technical factors and summarize the information from multiple probes to arrive at a single usable estimate of abundance or other relevant quantity, such as a ratio of two abundances, for each target. \end_layout \begin_layout Standard The choice of pre-processing algorithms used in the analysis of an array data set can have a large effect on the results of that analysis. However, despite their importance, these steps are often neglected or rushed in order to get to the more scientifically interesting analysis steps involving the actual biology of the system under study. Hence, it is often possible to achieve substantial gains in statistical power, model goodness-of-fit, or other relevant performance measures, by checking the assumptions made by each preprocessing step and choosing specific normalization methods tailored to the specific goals of the current analysis. \end_layout \begin_layout Subsection Normalization for clinical microarray classifiers must be single-channel \end_layout \begin_layout Subsubsection Standard normalization methods are unsuitable for clinical application \end_layout \begin_layout Standard As the cost of performing microarray assays falls, there is increasing interest in using genomic assays for diagnostic purposes, such as distinguishing healthy transplants (TX) from transplants undergoing acute rejection (AR) or acute dysfunction with no rejection (ADNR). However, the the standard normalization algorithm used for microarray data, Robust Multi-chip Average (RMA) \begin_inset CommandInset citation LatexCommand cite key "Irizarry2003a" literal "false" \end_inset , is not applicable in a clinical setting. Two of the steps in RMA, quantile normalization and probe summarization by median polish, depend on every array in the data set being normalized. This means that adding or removing any arrays from a data set changes the normalized values for all arrays, and data sets that have been normalized separately cannot be compared to each other. Hence, when using RMA, any arrays to be analyzed together must also be normalized together, and the set of arrays included in the data set must be held constant throughout an analysis. \end_layout \begin_layout Standard These limitations present serious impediments to the use of arrays as a diagnostic tool. When training a classifier, the samples to be classified must not be involved in any step of the training process, lest their inclusion bias the training process. Once a classifier is deployed in a clinical setting, the samples to be classified will not even \emph on exist \emph default at the time of training, so including them would be impossible even if it were statistically justifiable. Therefore, any machine learning application for microarrays demands that the normalized expression values computed for an array must depend only on information contained within that array. This would ensure that each array's normalization is independent of every other array, and that arrays normalized separately can still be compared to each other without bias. Such a normalization is commonly referred to as \begin_inset Quotes eld \end_inset single-channel normalization \begin_inset Quotes erd \end_inset . \end_layout \begin_layout Subsubsection Several strategies are available to meet clinical normalization requirements \end_layout \begin_layout Standard Frozen RMA (fRMA) addresses these concerns by replacing the quantile normalizati on and median polish with alternatives that do not introduce inter-array dependence, allowing each array to be normalized independently of all others \begin_inset CommandInset citation LatexCommand cite key "McCall2010" literal "false" \end_inset . Quantile normalization is performed against a pre-generated set of quantiles learned from a collection of 850 publically available arrays sampled from a wide variety of tissues in the Gene Expression Omnibus (GEO). Each array's probe intensity distribution is normalized against these pre-gener ated quantiles. The median polish step is replaced with a robust weighted average of probe intensities, using inverse variance weights learned from the same public GEO data. The result is a normalization that satisfies the requirements mentioned above: each array is normalized independently of all others, and any two normalized arrays can be compared directly to each other. \end_layout \begin_layout Standard One important limitation of fRMA is that it requires a separate reference data set from which to learn the parameters (reference quantiles and probe weights) that will be used to normalize each array. These parameters are specific to a given array platform, and pre-generated parameters are only provided for the most common platforms, such as Affymetrix hgu133plus2. For a less common platform, such as hthgu133pluspm, is is necessary to learn custom parameters from in-house data before fRMA can be used to normalize samples on that platform \begin_inset CommandInset citation LatexCommand cite key "McCall2011" literal "false" \end_inset . \end_layout \begin_layout Standard One other option is the aptly-named Single Channel Array Normalization (SCAN), which adapts a normalization method originally designed for tiling arrays \begin_inset CommandInset citation LatexCommand cite key "Piccolo2012" literal "false" \end_inset . SCAN is truly single-channel in that it does not require a set of normalization paramters estimated from an external set of reference samples like fRMA does. \end_layout \begin_layout Subsection Heteroskedasticity must be accounted for in methylation array data \end_layout \begin_layout Subsubsection Methylation array preprocessing induces heteroskedasticity \end_layout \begin_layout Standard DNA methylation arrays are a relatively new kind of assay that uses microarrays to measure the degree of methylation on cytosines in specific regions arrayed across the genome. First, bisulfite treatment converts all unmethylated cytosines to uracil (which then become thymine after amplication) while leaving methylated cytosines unaffected. Then, each target region is interrogated with two probes: one binds to the original genomic sequence and interrogates the level of methylated DNA, and the other binds to the sequence with all Cs replaced by Ts and interrogates the level of unmethylated DNA. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/methylvoom/sigmoid.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:Sigmoid-beta-m-mapping" \end_inset \series bold Sigmoid shape of the mapping between β and M values \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard After normalization, these two probe intensities are summarized in one of two ways, each with advantages and disadvantages. β \series bold \series default values, interpreted as fraction of DNA copies methylated, range from 0 to 1. β \series bold \series default values are conceptually easy to interpret, but the constrained range makes them unsuitable for linear modeling, and their error distributions are highly non-normal, which also frustrates linear modeling. M-values, interpreted as the log ratio of methylated to unmethylated copies, are computed by mapping the beta values from \begin_inset Formula $[0,1]$ \end_inset onto \begin_inset Formula $(-\infty,+\infty)$ \end_inset using a sigmoid curve (Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:Sigmoid-beta-m-mapping" plural "false" caps "false" noprefix "false" \end_inset ). This transformation results in values with better statistical perperties: the unconstrained range is suitable for linear modeling, and the error distributions are more normal. Hence, most linear modeling and other statistical testing on methylation arrays is performed using M-values. \end_layout \begin_layout Standard However, the steep slope of the sigmoid transformation near 0 and 1 tends to over-exaggerate small differences in β values near those extremes, which in turn amplifies the error in those values, leading to a U-shaped trend in the mean-variance curve: extreme values have higher variances than values near the middle. This mean-variance dependency must be accounted for when fitting the linear model for differential methylation, or else the variance will be systematically overestimated for probes with moderate M-values and underestimated for probes with extreme M-values. \end_layout \begin_layout Subsubsection The voom method for RNA-seq data can model M-value heteroskedasticity \end_layout \begin_layout Standard RNA-seq read count data are also known to show heteroskedasticity, and the voom method was developed for modeling this heteroskedasticity by estimating the mean-variance trend in the data and using this trend to assign precision weights to each observation \begin_inset CommandInset citation LatexCommand cite key "Law2013" literal "false" \end_inset . While methylation array data are not derived from counts and have a very different mean-variance relationship from that of typical RNA-seq data, the voom method makes no specific assumptions on the shape of the mean-variance relationship - it only assumes that the relationship is smooth enough to model using a lowess curve. Hence, the method is sufficiently general to model the mean-variance relationsh ip in methylation array data. However, the standard implementation of voom assumes that the input is given in raw read counts, and it must be adapted to run on methylation M-values. \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Put code on Github and reference it \end_layout \end_inset \end_layout \begin_layout Section Methods \end_layout \begin_layout Subsection Evaluation of classifier performance with different normalization methods \end_layout \begin_layout Standard For testing different normalizations, a data set of 157 hgu133plus2 arrays was used, consisting of blood samples from kidney transplant patients whose grafts had been graded as TX, AR, or ADNR via biopsy and histology (46 TX, 69 AR, 42 ADNR) \begin_inset CommandInset citation LatexCommand cite key "Kurian2014" literal "true" \end_inset . Additionally, an external validation set of 75 samples was gathered from public GEO data (37 TX, 38 AR, no ADNR). \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status collapsed \begin_layout Plain Layout Find appropriate GEO identifiers if possible. Kurian 2014 says GSE15296, but this seems to be different data. I also need to look up the GEO accession for the external validation set. \end_layout \end_inset \end_layout \begin_layout Standard To evaluate the effect of each normalization on classifier performance, the same classifier training and validation procedure was used after each normalization method. The PAM package was used to train a nearest shrunken centroid classifier on the training set and select the appropriate threshold for centroid shrinking. Then the trained classifier was used to predict the class probabilities of each validation sample. From these class probabilities, ROC curves and area-under-curve (AUC) values were generated \begin_inset CommandInset citation LatexCommand cite key "Turck2011" literal "false" \end_inset . Each normalization was tested on two different sets of training and validation samples. For internal validation, the 115 TX and AR arrays in the internal set were split at random into two equal sized sets, one for training and one for validation, each containing the same numbers of TX and AR samples as the other set. For external validation, the full set of 115 TX and AR samples were used as a training set, and the 75 external TX and AR samples were used as the validation set. Thus, 2 ROC curves and AUC values were generated for each normalization method: one internal and one external. Because the external validation set contains no ADNR samples, only classificati on of TX and AR samples was considered. The ADNR samples were included during normalization but excluded from all classifier training and validation. This ensures that the performance on internal and external validation sets is directly comparable. \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status collapsed \begin_layout Plain Layout Summarize the get.best.threshold algorithm for PAM threshold selection \end_layout \end_inset \end_layout \begin_layout Standard Six different normalization strategies were evaluated. First, 2 well-known non-single-channel normalization methods were considered: RMA and dChip \begin_inset CommandInset citation LatexCommand cite key "Li2001,Irizarry2003a" literal "false" \end_inset . Since RMA produces expression values on a log2 scale and dChip does not, the values from dChip were log2 transformed after normalization. Next, RMA and dChip followed by Global Rank-invariant Set Normalization (GRSN) were tested \begin_inset CommandInset citation LatexCommand cite key "Pelz2008" literal "false" \end_inset . Post-processing with GRSN does not turn RMA or dChip into single-channel methods, but it may help mitigate batch effects and is therefore useful as a benchmark. Lastly, the two single-channel normalization methods, fRMA and SCAN, were tested \begin_inset CommandInset citation LatexCommand cite key "McCall2010,Piccolo2012" literal "false" \end_inset . When evaluting internal validation performance, only the 157 internal samples were normalized; when evaluating external validation performance, all 157 internal samples and 75 external samples were normalized together. \end_layout \begin_layout Standard For demonstrating the problem with separate normalization of training and validation data, one additional normalization was performed: the internal and external sets were each normalized separately using RMA, and the normalized data for each set were combined into a single set with no further attempts at normalizing between the two sets. The represents approximately how RMA would have to be used in a clinical setting, where the samples to be classified are not available at the time the classifier is trained. \end_layout \begin_layout Subsection Generating custom fRMA vectors for hthgu133pluspm array platform \end_layout \begin_layout Standard In order to enable fRMA normalization for the hthgu133pluspm array platform, custom fRMA normalization vectors were trained using the frmaTools package \begin_inset CommandInset citation LatexCommand cite key "McCall2011" literal "false" \end_inset . Separate vectors were created for two types of samples: kidney graft biopsy samples and blood samples from graft recipients. For training, a 341 kidney biopsy samples from 2 data sets and 965 blood samples from 5 data sets were used as the reference set. Arrays were groups into batches based on unique combinations of sample type (blood or biopsy), diagnosis (TX, AR, etc.), data set, and scan date. Thus, each batch represents arrays of the same kind that were run together on the same day. For estimating the probe inverse variance weights, frmaTools requires equal-siz ed batches, which means a batch size must be chosen, and then batches smaller than that size must be ignored, while batches larger than the chosen size must be downsampled. This downsampling is performed randomly, so the sampling process is repeated 5 times and the resulting normalizations are compared to each other. \end_layout \begin_layout Standard To evaluate the consistency of the generated normalization vectors, the 5 fRMA vector sets generated from 5 random batch samplings were each used to normalize the same 20 randomly selected samples from each tissue. Then the normalized expression values for each probe on each array were compared across all normalizations. Each fRMA normalization was also compared against the normalized expression values obtained by normalizing the same 20 samples with ordinary RMA. \end_layout \begin_layout Subsection Modeling methylation array M-value heteroskedasticy with modified voom implement ation \end_layout \begin_layout Itemize Methylation arrays for differential methylation in rejection vs. healthy transplant \end_layout \begin_layout Itemize Adapt voom method originally designed for RNA-seq to model mean-variance dependence \end_layout \begin_layout Itemize Use sample precision weighting, duplicateCorrelation, and sva to adjust for other confounding factors \end_layout \begin_layout Section Results \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Improve subsection titles in this section \end_layout \end_inset \end_layout \begin_layout Subsection fRMA eliminates unwanted dependence of classifier training on normalization strategy caused by RMA \end_layout \begin_layout Subsubsection Separate normalization with RMA introduces unwanted biases in classification \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/PAM/predplot.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:Classifier-probabilities-RMA" \end_inset \series bold Classifier probabilities on validation samples when normalized with RMA together vs. separately. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard To demonstrate the problem with non-single-channel methods, we considered the problem of training a classifier to distinguish TX from AR using the samples from the internal set as training data, evaluating performance on the external set. First, training and evaluation were performed after normalizing all array samples together as a single set using RMA, and second, the internal samples were normalized separately from the external samples and the training and evaluation were repeated. For each sample in the validation set, the classifier probabilities from both classifiers were plotted against each other (Fig. \begin_inset CommandInset ref LatexCommand ref reference "fig:Classifier-probabilities-RMA" plural "false" caps "false" noprefix "false" \end_inset ). As expected, separate normalization biases the classifier probabilities, resulting in several misclassifications. In this case, the bias from separate normalization causes the classifier to assign a lower probability of AR to every sample. \end_layout \begin_layout Subsubsection fRMA and SCAN achieve maintain classification performance while eliminating dependence on normalization strategy \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/PAM/ROC-TXvsAR-internal.pdf width 100col% groupId colwidth \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:ROC-PAM-int" \end_inset ROC curves for PAM on internal validation data using different normalization strategies \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float table wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Tabular \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Normalization \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout Single-channel \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Internal Validation AUC \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout External Validation AUC \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none RMA \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout No \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.852 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.713 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none dChip \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout No \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.891 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.657 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none RMA + GRSN \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout No \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.816 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.750 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none dChip + GRSN \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout No \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.875 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.642 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none fRMA \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout Yes \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.863 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.718 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none SCAN \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout Yes \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.853 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0.689 \end_layout \end_inset \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "tab:AUC-PAM" \end_inset \series bold AUC values for internal and external validation with 6 different normalization strategies. \series default Only fRMA and SCAN are single-channel normalizations. The other 4 normalizations are for comparison. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard For internal validation, the 6 methods' AUC values ranged from 0.816 to 0.891, as shown in Table \begin_inset CommandInset ref LatexCommand ref reference "tab:AUC-PAM" plural "false" caps "false" noprefix "false" \end_inset . Among the non-single-channel normalizations, dChip outperformed RMA, while GRSN reduced the AUC values for both dChip and RMA. Both single-channel methods, fRMA and SCAN, slightly outperformed RMA, with fRMA ahead of SCAN. However, the difference between RMA and fRMA is still quite small. Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:ROC-PAM-int" plural "false" caps "false" noprefix "false" \end_inset shows that the ROC curves for RMA, dChip, and fRMA look very similar and relatively smooth, while both GRSN curves and the curve for SCAN have a more jagged appearance. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/PAM/ROC-TXvsAR-external.pdf width 100col% groupId colwidth \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:ROC-PAM-ext" \end_inset ROC curve for PAM on external validation data using different normalization strategies \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard For external validation, as expected, all the AUC values are lower than the internal validations, ranging from 0.642 to 0.750 (Table \begin_inset CommandInset ref LatexCommand ref reference "tab:AUC-PAM" plural "false" caps "false" noprefix "false" \end_inset ). With or without GRSN, RMA shows its dominance over dChip in this more challengi ng test. Unlike in the internal validation, GRSN actually improves the classifier performance for RMA, although it does not for dChip. Once again, both single-channel methods perform about on par with RMA, with fRMA performing slightly better and SCAN performing a bit worse. Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:ROC-PAM-ext" plural "false" caps "false" noprefix "false" \end_inset shows the ROC curves for the external validation test. As expected, none of them are as clean-looking as the internal validation ROC curves. The curves for RMA, RMA+GRSN, and fRMA all look similar, while the other curves look more divergent. \end_layout \begin_layout Subsection fRMA with custom-generated vectors enables normalization on hthgu133pluspm \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/batchsize_batches.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:batch-size-batches" \end_inset \series bold Effect of batch size selection on number of batches included in fRMA probe weight learning. \series default For batch sizes ranging from 3 to 15, the number of batches with at least that many samples was plotted for biopsy (BX) and blood (PAX) samples. The selected batch size, 5, is marked with a dotted vertical line. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/batchsize_samples.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:batch-size-samples" \end_inset \series bold Effect of batch size selection on number of samples included in fRMA probe weight learning. \series default For batch sizes ranging from 3 to 15, the number of samples included in probe weight training was plotted for biopsy (BX) and blood (PAX) samples. The selected batch size, 5, is marked with a dotted vertical line. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard In order to enable use of fRMA to normalize hthgu133pluspm, a custom set of fRMA vectors was created. First, an appropriate batch size was chosen by looking at the number of batches and number of samples included as a function of batch size (Figures \begin_inset CommandInset ref LatexCommand ref reference "fig:batch-size-batches" plural "false" caps "false" noprefix "false" \end_inset and \begin_inset CommandInset ref LatexCommand ref reference "fig:batch-size-samples" plural "false" caps "false" noprefix "false" \end_inset , respectively). For a given batch size, all batches with fewer samples that the chosen size must be ignored during training, while larger batches must be randomly downsampled to the chosen size. Hence, the number of samples included for a given batch size equals the batch size times the number of batches with at least that many samples. From Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:batch-size-samples" plural "false" caps "false" noprefix "false" \end_inset , it is apparent that that a batch size of 8 maximizes the number of samples included in training. Increasing the batch size beyond this causes too many smaller batches to be excluded, reducing the total number of samples for both tissue types. However, a batch size of 8 is not necessarily optimal. The article introducing frmaTools concluded that it was highly advantageous to use a smaller batch size in order to include more batches, even at the expense of including fewer total samples in training \begin_inset CommandInset citation LatexCommand cite key "McCall2011" literal "false" \end_inset . To strike an appropriate balance between more batches and more samples, a batch size of 5 was chosen. For both blood and biopsy samples, this increased the number of batches included by 10, with only a modest reduction in the number of samples compared to a batch size of 8. With a batch size of 5, 26 batches of biopsy samples and 46 batches of blood samples were available. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/M-BX-violin.pdf lyxscale 30 groupId m-violin \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:m-bx-violin" \end_inset \series bold Violin plot of log ratios between normalizations for 20 biopsy samples. \series default Each of 20 randomly selected biopsy samples was normalized with RMA and with 5 different sets of fRMA vectors. This shows the distribution of log ratios between normalized expression values, aggregated across all 20 arrays. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard Since fRMA training requires equal-size batches, larger batches are downsampled randomly. This introduces a nondeterministic step in the generation of normalization vectors. To show that this randomness does not substantially change the outcome, the random downsampling and subsequent vector learning was repeated 5 times, with a different random seed each time. 20 samples were selected at random as a test set and normalized with each of the 5 sets of fRMA normalization vectors as well as ordinary RMA, and the normalized expression values were compared across normalizations. Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:m-bx-violin" plural "false" caps "false" noprefix "false" \end_inset shows a summary of these comparisons for biopsy samples. Comparing RMA to each of the 5 fRMA normalizations, the distribution of log ratios is somewhat wide, indicating that the normalizations disagree on the expression values of a fair number of probe sets. In contrast, comparisons of fRMA against fRMA, the vast mojority of probe sets have very small log ratios, indicating a very high agreement between the normalized values generated by the two normalizations. This shows that the fRMA normalization's behavior is not very sensitive to the random downsampling of larger batches during training. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/MA-BX-RMA.fRMA.pdf lyxscale 50 groupId ma-frma \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:ma-bx-rma-frma" \end_inset \series bold Representative MA plot comparing RMA against fRMA for 20 biopsy samples. \series default Averages and log ratios were computed for every probe in each of 20 biopsy samples between RMA normalization and fRMA. Density of points is represented by darkness of shading, and individual outlier points are plotted. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/MA-BX-fRMA.fRMA.pdf lyxscale 50 groupId ma-frma \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:ma-bx-frma-frma" \end_inset \series bold Representative MA plot comparing different fRMA vectors for 20 biopsy samples. \series default Averages and log ratios were computed for every probe in each of 20 biopsy samples between fRMA normalizations using vectors from two different batch samplings. Density of points is represented by darkness of shading, and individual outlier points are plotted. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:ma-bx-rma-frma" plural "false" caps "false" noprefix "false" \end_inset shows an MA plot of the RMA-normalized values against the fRMA-normalized values for the same probe sets and arrays, corresponding to the first row of Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:m-bx-violin" plural "false" caps "false" noprefix "false" \end_inset . This MA plot shows that not only is there a wide distribution of M-values, but the trend of M-values is dependent on the average normalized intensity. This is expected, since the overall trend represents the differences in the quantile normalization step. When running RMA, only the quantiles for these specific 20 arrays are used, while for fRMA the quantile distribution is taking from all arrays used in training. Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:ma-bx-frma-frma" plural "false" caps "false" noprefix "false" \end_inset shows a similar MA plot comparing 2 different fRMA normalizations, correspondin g to the 6th row of Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:m-bx-violin" plural "false" caps "false" noprefix "false" \end_inset . The MA plot is very tightly centered around zero with no visible trend. Figures \begin_inset CommandInset ref LatexCommand ref reference "fig:m-pax-violin" plural "false" caps "false" noprefix "false" \end_inset , \begin_inset CommandInset ref LatexCommand ref reference "fig:MA-PAX-rma-frma" plural "false" caps "false" noprefix "false" \end_inset , and \begin_inset CommandInset ref LatexCommand ref reference "fig:ma-bx-frma-frma" plural "false" caps "false" noprefix "false" \end_inset show exactly the same information for the blood samples, once again comparing the normalized expression values between normalizations for all probe sets across 20 randomly selected test arrays. Once again, there is a wider distribution of log ratios between RMA-normalized values and fRMA-normalized, and a much tighter distribution when comparing different fRMA normalizations to each other, indicating that the fRMA training process is robust to random batch downsampling for the blood samples as well. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/M-PAX-violin.pdf lyxscale 30 groupId m-violin \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:m-pax-violin" \end_inset \series bold Violin plot of log ratios between normalizations for 20 blood samples. \series default Each of 20 randomly selected blood samples was normalized with RMA and with 5 different sets of fRMA vectors. This shows the distribution of log ratios between normalized expression values, aggregated across all 20 arrays. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/MA-PAX-RMA.fRMA.pdf lyxscale 50 groupId ma-frma \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:MA-PAX-rma-frma" \end_inset \series bold Representative MA plot comparing RMA against fRMA for 20 blood samples. \series default Averages and log ratios were computed for every probe in each of 20 blood samples between RMA normalization and fRMA. Density of points is represented by darkness of shading, and individual outlier points are plotted. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status collapsed \begin_layout Plain Layout \begin_inset Graphics filename graphics/frma-pax-bx/MA-PAX-fRMA.fRMA.pdf lyxscale 50 groupId ma-frma \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset CommandInset label LatexCommand label name "fig:MA-PAX-frma-frma" \end_inset \series bold Representative MA plot comparing different fRMA vectors for 20 blood samples. \series default Averages and log ratios were computed for every probe in each of 20 blood samples between fRMA normalizations using vectors from two different batch samplings. Density of points is represented by darkness of shading, and individual outlier points are plotted. \end_layout \end_inset \end_layout \end_inset \end_layout \begin_layout Subsection Adapting voom to methylation array data improves model fit \end_layout \begin_layout Itemize voom, precision weights, and sva improved model fit \end_layout \begin_deeper \begin_layout Itemize Also increased sensitivity for detecting differential methylation \end_layout \end_deeper \begin_layout Itemize Figure showing (a) heteroskedasticy without voom, (b) voom-modeled mean-variance trend, and (c) homoskedastic mean-variance trend after running voom \end_layout \begin_layout Itemize Figure showing sample weights and their relations to \end_layout \begin_layout Itemize Figure showing MDS plot with and without SVA correction \end_layout \begin_layout Itemize Figure and/or table showing improved p-value historgrams/number of significant genes (might need to get this from Padma) \end_layout \begin_layout Section Discussion \end_layout \begin_layout Subsection fRMA achieves clinically applicable normalization without sacrificing classifica tion performance \end_layout \begin_layout Standard As shown in Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:Classifier-probabilities-RMA" plural "false" caps "false" noprefix "false" \end_inset , improper normalization, particularly separate normalization of training and test samples, leads to unwanted biases in classification. In a controlled experimental context, it is always possible to correct this issue by normalizing all experimental samples together. However, because it is not feasible to normalize all samples together in a clinical context, a single-channel normalization is required is required. \end_layout \begin_layout Standard The major concern in using a single-channel normalization is that non-single-cha nnel methods can share information between arrays to improve the normalization, and single-channel methods risk sacrificing the gains in normalization accuracy that come from this information sharing. In the case of RMA, this information sharing is accomplished through quantile normalization and median polish steps. The need for information sharing in quantile normalization can easily be removed by learning a fixed set of quantiles from external data and normalizing each array to these fixed quantiles, instead of the quantiles of the data itself. As long as the fixed quantiles are reasonable, the result will be similar to standard RMA. However, there is no analogous way to eliminate cross-array information sharing in the median polish step, so fRMA replaces this with a weighted average of probes on each array, with the weights learned from external data. This step of fRMA has the greatest potential to diverge from RMA un undesirable ways. \end_layout \begin_layout Standard However, when run on real data, fRMA performed at least as well as RMA in both the internal validation and external validation tests. This shows that fRMA can be used to normalize individual clinical samples in a class prediction context without sacrificing the classifier performance that would be obtained by using the more well-established RMA for normalization. The other single-channel normalization method considered, SCAN, showed some loss of AUC in the external validation test. Based on these results, fRMA is the preferred normalization for clinical samples in a class prediction context. \end_layout \begin_layout Subsection Robust fRMA vectors can be generated for new array platforms \end_layout \begin_layout Standard The published fRMA normalization vectors for the hgu133plus2 platform were generated from a set of about 850 samples \begin_inset Flex TODO Note (Margin) status collapsed \begin_layout Plain Layout Look up the exact numbers \end_layout \end_inset chosen from a wide range of tissues, which the authors determined was sufficien t to generate a robust set of normalization vectors that could be applied across all tissues \begin_inset CommandInset citation LatexCommand cite key "McCall2010" literal "false" \end_inset . Since we only had hthgu133pluspm for 2 tissues of interest, our needs were more modest. Even using only 130 samples in 26 batches of 5 samples each for kidney biopsies, we were able to train a robust set of fRMA normalization vectors that were not meaningfully affected by the random selection of 5 samples from each batch. As expected, the training process was just as robust for the blood samples with 230 samples in 46 batches of 5 samples each. Because these vectors were each generated using training samples from a single tissue, they are not suitable for general use, unlike the vectors provided with fRMA itself. They are purpose-build for normalizing a specific type of sample on a specific platform. \end_layout \begin_layout Subsection voom \end_layout \begin_layout Itemize Methods like voom designed for RNA-seq can also help with array analysis \end_layout \begin_layout Itemize Extracting and modeling confounders common to many features improves model correspondence to known biology \end_layout \begin_layout Chapter Globin-blocking for more effective blood RNA-seq analysis in primate animal model \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Choose between above and the paper title: Optimizing yield of deep RNA sequencin g for gene expression profiling by globin reduction of peripheral blood samples from cynomolgus monkeys (Macaca fascicularis). \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Chapter author list: https://tex.stackexchange.com/questions/156862/displaying-aut hor-for-each-chapter-in-book Every chapter gets an author list, which may or may not be part of a citation to a published/preprinted paper. \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Preprint then cite the paper \end_layout \end_inset \end_layout \begin_layout Section* Abstract \end_layout \begin_layout Paragraph Background \end_layout \begin_layout Standard Primate blood contains high concentrations of globin messenger RNA. Globin reduction is a standard technique used to improve the expression results obtained by DNA microarrays on RNA from blood samples. However, with whole transcriptome RNA-sequencing (RNA-seq) quickly replacing microarrays for many applications, the impact of globin reduction for RNA-seq has not been previously studied. Moreover, no off-the-shelf kits are available for globin reduction in nonhuman primates. \end_layout \begin_layout Paragraph Results \end_layout \begin_layout Standard Here we report a protocol for RNA-seq in primate blood samples that uses complimentary oligonucleotides to block reverse transcription of the alpha and beta globin genes. In test samples from cynomolgus monkeys (Macaca fascicularis), this globin blocking protocol approximately doubles the yield of informative (non-globin) reads by greatly reducing the fraction of globin reads, while also improving the consistency in sequencing depth between samples. The increased yield enables detection of about 2000 more genes, significantly increases the correlation in measured gene expression levels between samples, and increases the sensitivity of differential gene expression tests. \end_layout \begin_layout Paragraph Conclusions \end_layout \begin_layout Standard These results show that globin blocking significantly improves the cost-effectiv eness of mRNA sequencing in primate blood samples by doubling the yield of useful reads, allowing detection of more genes, and improving the precision of gene expression measurements. Based on these results, a globin reducing or blocking protocol is recommended for all RNA-seq studies of primate blood samples. \end_layout \begin_layout Section Approach \end_layout \begin_layout Standard \begin_inset Note Note status open \begin_layout Plain Layout Consider putting some of this in the Intro chapter \end_layout \begin_layout Itemize Cynomolgus monkeys as a model organism \end_layout \begin_deeper \begin_layout Itemize Highly related to humans \end_layout \begin_layout Itemize Small size and short life cycle - good research animal \end_layout \begin_layout Itemize Genomics resources still in development \end_layout \end_deeper \begin_layout Itemize Inadequacy of existing blood RNA-seq protocols \end_layout \begin_deeper \begin_layout Itemize Existing protocols use a separate globin pulldown step, slowing down processing \end_layout \end_deeper \end_inset \end_layout \begin_layout Standard Increasingly, researchers are turning to high-throughput mRNA sequencing technologies (RNA-seq) in preference to expression microarrays for analysis of gene expression \begin_inset CommandInset citation LatexCommand cite key "Mutz2012" literal "false" \end_inset . The advantages are even greater for study of model organisms with no well-estab lished array platforms available, such as the cynomolgus monkey (Macaca fascicularis). High fractions of globin mRNA are naturally present in mammalian peripheral blood samples (up to 70% of total mRNA) and these are known to interfere with the results of array-based expression profiling \begin_inset CommandInset citation LatexCommand cite key "Winn2010" literal "false" \end_inset . The importance of globin reduction for RNA-seq of blood has only been evaluated for a deepSAGE protocol on human samples \begin_inset CommandInset citation LatexCommand cite key "Mastrokolias2012" literal "false" \end_inset . In the present report, we evaluated globin reduction using custom blocking oligonucleotides for deep RNA-seq of peripheral blood samples from a nonhuman primate, cynomolgus monkey, using the Illumina technology platform. We demonstrate that globin reduction significantly improves the cost-effectiven ess of RNA-seq in blood samples. Thus, our protocol offers a significant advantage to any investigator planning to use RNA-seq for gene expression profiling of nonhuman primate blood samples. Our method can be generally applied to any species by designing complementary oligonucleotide blocking probes to the globin gene sequences of that species. Indeed, any highly expressed but biologically uninformative transcripts can also be blocked to further increase sequencing efficiency and value \begin_inset CommandInset citation LatexCommand cite key "Arnaud2016" literal "false" \end_inset . \end_layout \begin_layout Section Methods \end_layout \begin_layout Subsection* Sample collection \end_layout \begin_layout Standard All research reported here was done under IACUC-approved protocols at the University of Miami and complied with all applicable federal and state regulations and ethical principles for nonhuman primate research. Blood draws occurred between 16 April 2012 and 18 June 2015. The experimental system involved intrahepatic pancreatic islet transplantation into Cynomolgus monkeys with induced diabetes mellitus with or without concomitant infusion of mesenchymal stem cells. Blood was collected at serial time points before and after transplantation into PAXgene Blood RNA tubes (PreAnalytiX/Qiagen, Valencia, CA) at the precise volume:volume ratio of 2.5 ml whole blood into 6.9 ml of PAX gene additive. \end_layout \begin_layout Subsection* Globin Blocking \end_layout \begin_layout Standard Four oligonucleotides were designed to hybridize to the 3’ end of the transcript s for Cynomolgus HBA1, HBA2 and HBB, with two hybridization sites for HBB and 2 sites for HBA (the chosen sites were identical in both HBA genes). All oligos were purchased from Sigma and were entirely composed of 2’O-Me bases with a C3 spacer positioned at the 3’ ends to prevent any polymerase mediated primer extension. \end_layout \begin_layout Quote HBA1/2 site 1: GCCCACUCAGACUUUAUUCAAAG-C3spacer \end_layout \begin_layout Quote HBA1/2 site 2: GGUGCAAGGAGGGGAGGAG-C3spacer \end_layout \begin_layout Quote HBB site 1: AAUGAAAAUAAAUGUUUUUUAUUAG-C3spacer \end_layout \begin_layout Quote HBB site 2: CUCAAGGCCCUUCAUAAUAUCCC-C3spacer \end_layout \begin_layout Subsection* RNA-seq Library Preparation \end_layout \begin_layout Standard Sequencing libraries were prepared with 200ng total RNA from each sample. Polyadenylated mRNA was selected from 200 ng aliquots of cynomologus blood-deri ved total RNA using Ambion Dynabeads Oligo(dT)25 beads (Invitrogen) following manufacturer’s recommended protocol. PolyA selected RNA was then combined with 8 pmol of HBA1/2 (site 1), 8 pmol of HBA1/2 (site 2), 12 pmol of HBB (site 1) and 12 pmol of HBB (site 2) oligonucleotides. In addition, 20 pmol of RT primer containing a portion of the Illumina adapter sequence (B-oligo-dTV: GAGTTCCTTGGCACCCGAGAATTCCATTTTTTTTTTTTTTTTTTTV) and 4 µL of 5X First Strand buffer (250 mM Tris-HCl pH 8.3, 375 mM KCl, 15mM MgCl2) were added in a total volume of 15 µL. The RNA was fragmented by heating this cocktail for 3 minutes at 95°C and then placed on ice. This was followed by the addition of 2 µL 0.1 M DTT, 1 µL RNaseOUT, 1 µL 10mM dNTPs 10% biotin-16 aminoallyl-2’- dUTP and 10% biotin-16 aminoallyl-2’- dCTP (TriLink Biotech, San Diego, CA), 1 µL Superscript II (200U/ µL, Thermo-Fi sher). A second “unblocked” library was prepared in the same way for each sample but replacing the blocking oligos with an equivalent volume of water. The reaction was carried out at 25°C for 15 minutes and 42°C for 40 minutes, followed by incubation at 75°C for 10 minutes to inactivate the reverse transcriptase. \end_layout \begin_layout Standard The cDNA/RNA hybrid molecules were purified using 1.8X Ampure XP beads (Agencourt ) following supplier’s recommended protocol. The cDNA/RNA hybrid was eluted in 25 µL of 10 mM Tris-HCl pH 8.0, and then bound to 25 µL of M280 Magnetic Streptavidin beads washed per recommended protocol (Thermo-Fisher). After 30 minutes of binding, beads were washed one time in 100 µL 0.1N NaOH to denature and remove the bound RNA, followed by two 100 µL washes with 1X TE buffer. \end_layout \begin_layout Standard Subsequent attachment of the 5-prime Illumina A adapter was performed by on-bead random primer extension of the following sequence (A-N8 primer: TTCAGAGTTCTACAGTCCGACGATCNNNNNNNN). Briefly, beads were resuspended in a 20 µL reaction containing 5 µM A-N8 primer, 40mM Tris-HCl pH 7.5, 20mM MgCl2, 50mM NaCl, 0.325U/µL Sequenase 2.0 (Affymetrix, Santa Clara, CA), 0.0025U/µL inorganic pyrophosphatase (Affymetr ix) and 300 µM each dNTP. Reaction was incubated at 22°C for 30 minutes, then beads were washed 2 times with 1X TE buffer (200µL). \end_layout \begin_layout Standard The magnetic streptavidin beads were resuspended in 34 µL nuclease-free water and added directly to a PCR tube. The two Illumina protocol-specified PCR primers were added at 0.53 µM (Illumina TruSeq Universal Primer 1 and Illumina TruSeq barcoded PCR primer 2), along with 40 µL 2X KAPA HiFi Hotstart ReadyMix (KAPA, Willmington MA) and thermocycl ed as follows: starting with 98°C (2 min-hold); 15 cycles of 98°C, 20sec; 60°C, 30sec; 72°C, 30sec; and finished with a 72°C (2 min-hold). \end_layout \begin_layout Standard PCR products were purified with 1X Ampure Beads following manufacturer’s recommended protocol. Libraries were then analyzed using the Agilent TapeStation and quantitation of desired size range was performed by “smear analysis”. Samples were pooled in equimolar batches of 16 samples. Pooled libraries were size selected on 2% agarose gels (E-Gel EX Agarose Gels; Thermo-Fisher). Products were cut between 250 and 350 bp (corresponding to insert sizes of 130 to 230 bps). Finished library pools were then sequenced on the Illumina NextSeq500 instrumen t with 75 base read lengths. \end_layout \begin_layout Subsection* Read alignment and counting \end_layout \begin_layout Standard Reads were aligned to the cynomolgus genome using STAR \begin_inset CommandInset citation LatexCommand cite key "Dobin2013,Wilson2013" literal "false" \end_inset . Counts of uniquely mapped reads were obtained for every gene in each sample with the “featureCounts” function from the Rsubread package, using each of the three possibilities for the “strandSpecific” option: sense, antisense, and unstranded \begin_inset CommandInset citation LatexCommand cite key "Liao2014" literal "false" \end_inset . A few artifacts in the cynomolgus genome annotation complicated read counting. First, no ortholog is annotated for alpha globin in the cynomolgus genome, presumably because the human genome has two alpha globin genes with nearly identical sequences, making the orthology relationship ambiguous. However, two loci in the cynomolgus genome are as “hemoglobin subunit alpha-lik e” (LOC102136192 and LOC102136846). LOC102136192 is annotated as a pseudogene while LOC102136846 is annotated as protein-coding. Our globin reduction protocol was designed to include blocking of these two genes. Indeed, these two genes have almost the same read counts in each library as the properly-annotated HBB gene and much larger counts than any other gene in the unblocked libraries, giving confidence that reads derived from the real alpha globin are mapping to both genes. Thus, reads from both of these loci were counted as alpha globin reads in all further analyses. The second artifact is a small, uncharacterized non-coding RNA gene (LOC1021365 91), which overlaps the HBA-like gene (LOC102136192) on the opposite strand. If counting is not performed in stranded mode (or if a non-strand-specific sequencing protocol is used), many reads mapping to the globin gene will be discarded as ambiguous due to their overlap with this ncRNA gene, resulting in significant undercounting of globin reads. Therefore, stranded sense counts were used for all further analysis in the present study to insure that we accurately accounted for globin transcript reduction. However, we note that stranded reads are not necessary for RNA-seq using our protocol in standard practice. \end_layout \begin_layout Subsection* Normalization and Exploratory Data Analysis \end_layout \begin_layout Standard Libraries were normalized by computing scaling factors using the edgeR package’s Trimmed Mean of M-values method \begin_inset CommandInset citation LatexCommand cite key "Robinson2010" literal "false" \end_inset . Log2 counts per million values (logCPM) were calculated using the cpm function in edgeR for individual samples and aveLogCPM function for averages across groups of samples, using those functions’ default prior count values to avoid taking the logarithm of 0. Genes were considered “present” if their average normalized logCPM values across all libraries were at least -1. Normalizing for gene length was unnecessary because the sequencing protocol is 3’-biased and hence the expected read count for each gene is related to the transcript’s copy number but not its length. \end_layout \begin_layout Standard In order to assess the effect of blocking on reproducibility, Pearson and Spearman correlation coefficients were computed between the logCPM values for every pair of libraries within the globin-blocked (GB) and unblocked (non-GB) groups, and edgeR's “estimateDisp” function was used to compute negative binomial dispersions separately for the two groups \begin_inset CommandInset citation LatexCommand cite key "Chen2014" literal "false" \end_inset . \end_layout \begin_layout Subsection* Differential Expression Analysis \end_layout \begin_layout Standard All tests for differential gene expression were performed using edgeR, by first fitting a negative binomial generalized linear model to the counts and normalization factors and then performing a quasi-likelihood F-test with robust estimation of outlier gene dispersions \begin_inset CommandInset citation LatexCommand cite key "Lund2012,Phipson2016" literal "false" \end_inset . To investigate the effects of globin blocking on each gene, an additive model was fit to the full data with coefficients for globin blocking and SampleID. To test the effect of globin blocking on detection of differentially expressed genes, the GB samples and non-GB samples were each analyzed independently as follows: for each animal with both a pre-transplant and a post-transplant time point in the data set, the pre-transplant sample and the earliest post-transplant sample were selected, and all others were excluded, yielding a pre-/post-transplant pair of samples for each animal (N=7 animals with paired samples). These samples were analyzed for pre-transplant vs. post-transplant differential gene expression while controlling for inter-animal variation using an additive model with coefficients for transplant and animal ID. In all analyses, p-values were adjusted using the Benjamini-Hochberg procedure for FDR correction \begin_inset CommandInset citation LatexCommand cite key "Benjamini1995" literal "false" \end_inset . \end_layout \begin_layout Standard \begin_inset Note Note status open \begin_layout Itemize New blood RNA-seq protocol to block reverse transcription of globin genes \end_layout \begin_layout Itemize Blood RNA-seq time course after transplants with/without MSC infusion \end_layout \end_inset \end_layout \begin_layout Section Results \end_layout \begin_layout Subsection* Globin blocking yields a larger and more consistent fraction of useful reads \end_layout \begin_layout Standard The objective of the present study was to validate a new protocol for deep RNA-seq of whole blood drawn into PaxGene tubes from cynomolgus monkeys undergoing islet transplantation, with particular focus on minimizing the loss of useful sequencing space to uninformative globin reads. The details of the analysis with respect to transplant outcomes and the impact of mesenchymal stem cell treatment will be reported in a separate manuscript (in preparation). To focus on the efficacy of our globin blocking protocol, 37 blood samples, 16 from pre-transplant and 21 from post-transplant time points, were each prepped once with and once without globin blocking oligos, and were then sequenced on an Illumina NextSeq500 instrument. The number of reads aligning to each gene in the cynomolgus genome was counted. Table 1 summarizes the distribution of read fractions among the GB and non-GB libraries. In the libraries with no globin blocking, globin reads made up an average of 44.6% of total input reads, while reads assigned to all other genes made up an average of 26.3%. The remaining reads either aligned to intergenic regions (that include long non-coding RNAs) or did not align with any annotated transcripts in the current build of the cynomolgus genome. In the GB libraries, globin reads made up only 3.48% and reads assigned to all other genes increased to 50.4%. Thus, globin blocking resulted in a 92.2% reduction in globin reads and a 91.6% increase in yield of useful non-globin reads. \end_layout \begin_layout Standard This reduction is not quite as efficient as the previous analysis showed for human samples by DeepSAGE (<0.4% globin reads after globin reduction) \begin_inset CommandInset citation LatexCommand cite key "Mastrokolias2012" literal "false" \end_inset . Nonetheless, this degree of globin reduction is sufficient to nearly double the yield of useful reads. Thus, globin blocking cuts the required sequencing effort (and costs) to achieve a target coverage depth by almost 50%. Consistent with this near doubling of yield, the average difference in un-normalized logCPM across all genes between the GB libraries and non-GB libraries is approximately 1 (mean = 1.01, median = 1.08), an overall 2-fold increase. Un-normalized values are used here because the TMM normalization correctly identifies this 2-fold difference as biologically irrelevant and removes it. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \align center \begin_inset Graphics filename graphics/Globin Paper/figure1 - globin-fractions.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \series bold \begin_inset Argument 1 status collapsed \begin_layout Plain Layout Fraction of genic reads in each sample aligned to non-globin genes, with and without globin blocking (GB). \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "fig:Fraction-of-genic-reads" \end_inset Fraction of genic reads in each sample aligned to non-globin genes, with and without globin blocking (GB). \series default All reads in each sequencing library were aligned to the cyno genome, and the number of reads uniquely aligning to each gene was counted. For each sample, counts were summed separately for all globin genes and for the remainder of the genes (non-globin genes), and the fraction of genic reads aligned to non-globin genes was computed. Each point represents an individual sample. Gray + signs indicate the means for globin-blocked libraries and unblocked libraries. The overall distribution for each group is represented as a notched box plots. Points are randomly spread vertically to avoid excessive overlapping. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float table placement p wide false sideways true status open \begin_layout Plain Layout \align center \begin_inset Tabular \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Percent of Total Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Percent of Genic Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout GB \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Non-globin Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Globin Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none All Genic Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none All Aligned Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Non-globin Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Globin Reads \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none Yes \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 50.4% ± 6.82 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 3.48% ± 2.94 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 53.9% ± 6.81 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 89.7% ± 2.40 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 93.5% ± 5.25 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 6.49% ± 5.25 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none No \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 26.3% ± 8.95 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 44.6% ± 16.6 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 70.1% ± 9.38 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 90.7% ± 5.16 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 38.8% ± 17.1 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 61.2% ± 17.1 \end_layout \end_inset \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \series bold \begin_inset Argument 1 status collapsed \begin_layout Plain Layout Fractions of reads mapping to genomic features in GB and non-GB samples. \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "tab:Fractions-of-reads" \end_inset Fractions of reads mapping to genomic features in GB and non-GB samples. \series default All values are given as mean ± standard deviation. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard Another important aspect is that the standard deviations in Table \begin_inset CommandInset ref LatexCommand ref reference "tab:Fractions-of-reads" plural "false" caps "false" noprefix "false" \end_inset are uniformly smaller in the GB samples than the non-GB ones, indicating much greater consistency of yield. This is best seen in the percentage of non-globin reads as a fraction of total reads aligned to annotated genes (genic reads). For the non-GB samples, this measure ranges from 10.9% to 80.9%, while for the GB samples it ranges from 81.9% to 99.9% (Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:Fraction-of-genic-reads" plural "false" caps "false" noprefix "false" \end_inset ). This means that for applications where it is critical that each sample achieve a specified minimum coverage in order to provide useful information, it would be necessary to budget up to 10 times the sequencing depth per sample without globin blocking, even though the average yield improvement for globin blocking is only 2-fold, because every sample has a chance of being 90% globin and 10% useful reads. Hence, the more consistent behavior of GB samples makes planning an experiment easier and more efficient because it eliminates the need to over-sequence every sample in order to guard against the worst case of a high-globin fraction. \end_layout \begin_layout Subsection* Globin blocking lowers the noise floor and allows detection of about 2000 more genes \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Remove redundant titles from figures \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \align center \begin_inset Graphics filename graphics/Globin Paper/figure2 - aveLogCPM-colored.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \series bold \begin_inset Argument 1 status collapsed \begin_layout Plain Layout Distributions of average group gene abundances when normalized separately or together. \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "fig:logcpm-dists" \end_inset Distributions of average group gene abundances when normalized separately or together. \series default All reads in each sequencing library were aligned to the cyno genome, and the number of reads uniquely aligning to each gene was counted. Genes with zero counts in all libraries were discarded. Libraries were normalized using the TMM method. Libraries were split into globin-blocked (GB) and non-GB groups and the average abundance for each gene in both groups, measured in log2 counts per million reads counted, was computed using the aveLogCPM function. The distribution of average gene logCPM values was plotted for both groups using a kernel density plot to approximate a continuous distribution. The logCPM GB distributions are marked in red, non-GB in blue. The black vertical line denotes the chosen detection threshold of -1. Top panel: Libraries were split into GB and non-GB groups first and normalized separately. Bottom panel: Libraries were all normalized together first and then split into groups. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard Since globin blocking yields more usable sequencing depth, it should also allow detection of more genes at any given threshold. When we looked at the distribution of average normalized logCPM values across all libraries for genes with at least one read assigned to them, we observed the expected bimodal distribution, with a high-abundance "signal" peak representing detected genes and a low-abundance "noise" peak representing genes whose read count did not rise above the noise floor (Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:logcpm-dists" plural "false" caps "false" noprefix "false" \end_inset ). Consistent with the 2-fold increase in raw counts assigned to non-globin genes, the signal peak for GB samples is shifted to the right relative to the non-GB signal peak. When all the samples are normalized together, this difference is normalized out, lining up the signal peaks, and this reveals that, as expected, the noise floor for the GB samples is about 2-fold lower. This greater separation between signal and noise peaks in the GB samples means that low-expression genes should be more easily detected and more precisely quantified than in the non-GB samples. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \align center \begin_inset Graphics filename graphics/Globin Paper/figure3 - detection.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \series bold \begin_inset Argument 1 status collapsed \begin_layout Plain Layout Gene detections as a function of abundance thresholds in globin-blocked (GB) and non-GB samples. \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "fig:Gene-detections" \end_inset Gene detections as a function of abundance thresholds in globin-blocked (GB) and non-GB samples. \series default Average abundance (logCPM, \begin_inset Formula $\log_{2}$ \end_inset counts per million reads counted) was computed by separate group normalization as described in Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:logcpm-dists" plural "false" caps "false" noprefix "false" \end_inset for both the GB and non-GB groups, as well as for all samples considered as one large group. For each every integer threshold from -2 to 3, the number of genes detected at or above that logCPM threshold was plotted for each group. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard Based on these distributions, we selected a detection threshold of -1, which is approximately the leftmost edge of the trough between the signal and noise peaks. This represents the most liberal possible detection threshold that doesn't call substantial numbers of noise genes as detected. Among the full dataset, 13429 genes were detected at this threshold, and 22276 were not. When considering the GB libraries and non-GB libraries separately and re-comput ing normalization factors independently within each group, 14535 genes were detected in the GB libraries while only 12460 were detected in the non-GB libraries. Thus, GB allowed the detection of 2000 extra genes that were buried under the noise floor without GB. This pattern of at least 2000 additional genes detected with GB was also consistent across a wide range of possible detection thresholds, from -2 to 3 (see Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:Gene-detections" plural "false" caps "false" noprefix "false" \end_inset ). \end_layout \begin_layout Subsection* Globin blocking does not add significant additional noise or decrease sample quality \end_layout \begin_layout Standard One potential worry is that the globin blocking protocol could perturb the levels of non-globin genes. There are two kinds of possible perturbations: systematic and random. The former is not a major concern for detection of differential expression, since a 2-fold change in every sample has no effect on the relative fold change between samples. In contrast, random perturbations would increase the noise and obscure the signal in the dataset, reducing the capacity to detect differential expression. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \align center \begin_inset Graphics filename graphics/Globin Paper/figure4 - maplot-colored.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \begin_inset Argument 1 status collapsed \begin_layout Plain Layout MA plot showing effects of globin blocking on each gene's abundance. \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "fig:MA-plot" \end_inset \series bold MA plot showing effects of globin blocking on each gene's abundance. \series default All libraries were normalized together as described in Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:logcpm-dists" plural "false" caps "false" noprefix "false" \end_inset , and genes with an average logCPM below -1 were filtered out. Each remaining gene was tested for differential abundance with respect to globin blocking (GB) using edgeR’s quasi-likelihod F-test, fitting a negative binomial generalized linear model to table of read counts in each library. For each gene, edgeR reported average abundance (logCPM), \begin_inset Formula $\log_{2}$ \end_inset fold change (logFC), p-value, and Benjamini-Hochberg adjusted false discovery rate (FDR). Each gene's logFC was plotted against its logCPM, colored by FDR. Red points are significant at ≤10% FDR, and blue are not significant at that threshold. The alpha and beta globin genes targeted for blocking are marked with large triangles, while all other genes are represented as small points. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Standardize on \begin_inset Quotes eld \end_inset log2 \begin_inset Quotes erd \end_inset notation \end_layout \end_inset \end_layout \begin_layout Standard The data do indeed show small systematic perturbations in gene levels (Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:MA-plot" plural "false" caps "false" noprefix "false" \end_inset ). Other than the 3 designated alpha and beta globin genes, two other genes stand out as having especially large negative log fold changes: HBD and LOC1021365. HBD, delta globin, is most likely targeted by the blocking oligos due to high sequence homology with the other globin genes. LOC1021365 is the aforementioned ncRNA that is reverse-complementary to one of the alpha-like genes and that would be expected to be removed during the globin blocking step. All other genes appear in a cluster centered vertically at 0, and the vast majority of genes in this cluster show an absolute log2(FC) of 0.5 or less. Nevertheless, many of these small perturbations are still statistically significant, indicating that the globin blocking oligos likely cause very small but non-zero systematic perturbations in measured gene expression levels. \end_layout \begin_layout Standard \begin_inset Float figure wide false sideways false status open \begin_layout Plain Layout \align center \begin_inset Graphics filename graphics/Globin Paper/figure5 - corrplot.pdf \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \series bold \begin_inset Argument 1 status collapsed \begin_layout Plain Layout Comparison of inter-sample gene abundance correlations with and without globin blocking. \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "fig:gene-abundance-correlations" \end_inset Comparison of inter-sample gene abundance correlations with and without globin blocking (GB). \series default All libraries were normalized together as described in Figure 2, and genes with an average abundance (logCPM, log2 counts per million reads counted) less than -1 were filtered out. Each gene’s logCPM was computed in each library using the edgeR cpm function. For each pair of biological samples, the Pearson correlation between those samples' GB libraries was plotted against the correlation between the same samples’ non-GB libraries. Each point represents an unique pair of samples. The solid gray line shows a quantile-quantile plot of distribution of GB correlations vs. that of non-GB correlations. The thin dashed line is the identity line, provided for reference. \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard To evaluate the possibility of globin blocking causing random perturbations and reducing sample quality, we computed the Pearson correlation between logCPM values for every pair of samples with and without GB and plotted them against each other (Figure \begin_inset CommandInset ref LatexCommand ref reference "fig:gene-abundance-correlations" plural "false" caps "false" noprefix "false" \end_inset ). The plot indicated that the GB libraries have higher sample-to-sample correlati ons than the non-GB libraries. Parametric and nonparametric tests for differences between the correlations with and without GB both confirmed that this difference was highly significant (2-sided paired t-test: t = 37.2, df = 665, P ≪ 2.2e-16; 2-sided Wilcoxon sign-rank test: V = 2195, P ≪ 2.2e-16). Performing the same tests on the Spearman correlations gave the same conclusion (t-test: t = 26.8, df = 665, P ≪ 2.2e-16; sign-rank test: V = 8781, P ≪ 2.2e-16). The edgeR package was used to compute the overall biological coefficient of variation (BCV) for GB and non-GB libraries, and found that globin blocking resulted in a negligible increase in the BCV (0.417 with GB vs. 0.400 without). The near equality of the BCVs for both sets indicates that the higher correlati ons in the GB libraries are most likely a result of the increased yield of useful reads, which reduces the contribution of Poisson counting uncertainty to the overall variance of the logCPM values \begin_inset CommandInset citation LatexCommand cite key "McCarthy2012" literal "false" \end_inset . This improves the precision of expression measurements and more than offsets the negligible increase in BCV. \end_layout \begin_layout Subsection* More differentially expressed genes are detected with globin blocking \end_layout \begin_layout Standard \begin_inset Float table wide false sideways false status open \begin_layout Plain Layout \align center \begin_inset Tabular \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold No Globin Blocking \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold Up \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold NS \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold Down \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold Globin-Blocking \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold Up \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 231 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 515 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 2 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold NS \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 160 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 11235 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 136 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \series bold Down \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 0 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 548 \end_layout \end_inset \begin_inset Text \begin_layout Plain Layout \family roman \series medium \shape up \size normal \emph off \bar no \strikeout off \xout off \uuline off \uwave off \noun off \color none 127 \end_layout \end_inset \end_inset \end_layout \begin_layout Plain Layout \begin_inset Caption Standard \begin_layout Plain Layout \series bold \begin_inset Argument 1 status open \begin_layout Plain Layout Comparison of significantly differentially expressed genes with and without globin blocking. \end_layout \end_inset \begin_inset CommandInset label LatexCommand label name "tab:Comparison-of-significant" \end_inset Comparison of significantly differentially expressed genes with and without globin blocking. \series default Up, Down: Genes significantly up/down-regulated in post-transplant samples relative to pre-transplant samples, with a false discovery rate of 10% or less. NS: Non-significant genes (false discovery rate greater than 10%). \end_layout \end_inset \end_layout \begin_layout Plain Layout \end_layout \end_inset \end_layout \begin_layout Standard To compare performance on differential gene expression tests, we took subsets of both the GB and non-GB libraries with exactly one pre-transplant and one post-transplant sample for each animal that had paired samples available for analysis (N=7 animals, N=14 samples in each subset). The same test for pre- vs. post-transplant differential gene expression was performed on the same 7 pairs of samples from GB libraries and non-GB libraries, in each case using an FDR of 10% as the threshold of significance. Out of 12954 genes that passed the detection threshold in both subsets, 358 were called significantly differentially expressed in the same direction in both sets; 1063 were differentially expressed in the GB set only; 296 were differentially expressed in the non-GB set only; 2 genes were called significantly up in the GB set but significantly down in the non-GB set; and the remaining 11235 were not called differentially expressed in either set. These data are summarized in Table \begin_inset CommandInset ref LatexCommand ref reference "tab:Comparison-of-significant" plural "false" caps "false" noprefix "false" \end_inset . The differences in BCV calculated by EdgeR for these subsets of samples were negligible (BCV = 0.302 for GB and 0.297 for non-GB). \end_layout \begin_layout Standard The key point is that the GB data results in substantially more differentially expressed calls than the non-GB data. Since there is no gold standard for this dataset, it is impossible to be certain whether this is due to under-calling of differential expression in the non-GB samples or over-calling in the GB samples. However, given that both datasets are derived from the same biological samples and have nearly equal BCVs, it is more likely that the larger number of DE calls in the GB samples are genuine detections that were enabled by the higher sequencing depth and measurement precision of the GB samples. Note that the same set of genes was considered in both subsets, so the larger number of differentially expressed gene calls in the GB data set reflects a greater sensitivity to detect significant differential gene expression and not simply the larger total number of detected genes in GB samples described earlier. \end_layout \begin_layout Section Discussion \end_layout \begin_layout Standard The original experience with whole blood gene expression profiling on DNA microarrays demonstrated that the high concentration of globin transcripts reduced the sensitivity to detect genes with relatively low expression levels, in effect, significantly reducing the sensitivity. To address this limitation, commercial protocols for globin reduction were developed based on strategies to block globin transcript amplification during labeling or physically removing globin transcripts by affinity bead methods \begin_inset CommandInset citation LatexCommand cite key "Winn2010" literal "false" \end_inset . More recently, using the latest generation of labeling protocols and arrays, it was determined that globin reduction was no longer necessary to obtain sufficient sensitivity to detect differential transcript expression \begin_inset CommandInset citation LatexCommand cite key "NuGEN2010" literal "false" \end_inset . However, we are not aware of any publications using these currently available protocols the with latest generation of microarrays that actually compare the detection sensitivity with and without globin reduction. However, in practice this has now been adopted generally primarily driven by concerns for cost control. The main objective of our work was to directly test the impact of globin gene transcripts and a new globin blocking protocol for application to the newest generation of differential gene expression profiling determined using next generation sequencing. \end_layout \begin_layout Standard The challenge of doing global gene expression profiling in cynomolgus monkeys is that the current available arrays were never designed to comprehensively cover this genome and have not been updated since the first assemblies of the cynomolgus genome were published. Therefore, we determined that the best strategy for peripheral blood profiling was to do deep RNA-seq and inform the workflow using the latest available genome assembly and annotation \begin_inset CommandInset citation LatexCommand cite key "Wilson2013" literal "false" \end_inset . However, it was not immediately clear whether globin reduction was necessary for RNA-seq or how much improvement in efficiency or sensitivity to detect differential gene expression would be achieved for the added cost and work. \end_layout \begin_layout Standard We only found one report that demonstrated that globin reduction significantly improved the effective read yields for sequencing of human peripheral blood cell RNA using a DeepSAGE protocol \begin_inset CommandInset citation LatexCommand cite key "Mastrokolias2012" literal "false" \end_inset . The approach to DeepSAGE involves two different restriction enzymes that purify and then tag small fragments of transcripts at specific locations and thus, significantly reduces the complexity of the transcriptome. Therefore, we could not determine how DeepSAGE results would translate to the common strategy in the field for assaying the entire transcript population by whole-transcriptome 3’-end RNA-seq. Furthermore, if globin reduction is necessary, we also needed a globin reduction method specific to cynomolgus globin sequences that would work an organism for which no kit is available off the shelf. \end_layout \begin_layout Standard As mentioned above, the addition of globin blocking oligos has a very small impact on measured expression levels of gene expression. However, this is a non-issue for the purposes of differential expression testing, since a systematic change in a gene in all samples does not affect relative expression levels between samples. However, we must acknowledge that simple comparisons of gene expression data obtained by GB and non-GB protocols are not possible without additional normalization. \end_layout \begin_layout Standard More importantly, globin blocking not only nearly doubles the yield of usable reads, it also increases inter-sample correlation and sensitivity to detect differential gene expression relative to the same set of samples profiled without blocking. In addition, globin blocking does not add a significant amount of random noise to the data. Globin blocking thus represents a cost-effective way to squeeze more data and statistical power out of the same blood samples and the same amount of sequencing. In conclusion, globin reduction greatly increases the yield of useful RNA-seq reads mapping to the rest of the genome, with minimal perturbations in the relative levels of non-globin genes. Based on these results, globin transcript reduction using sequence-specific, complementary blocking oligonucleotides is recommended for all deep RNA-seq of cynomolgus and other nonhuman primate blood samples. \end_layout \begin_layout Chapter Future Directions \end_layout \begin_layout Itemize Study other epigenetic marks in more contexts \end_layout \begin_deeper \begin_layout Itemize DNA methylation, histone marks, chromatin accessibility & conformation in CD4 T-cells \end_layout \begin_layout Itemize Also look at other types lymphocytes: CD8 T-cells, B-cells, NK cells \end_layout \end_deeper \begin_layout Itemize Use CV or bootstrap to better evaluate classifiers \end_layout \begin_layout Standard \begin_inset ERT status open \begin_layout Plain Layout % Use "References" instead of "Bibliography" \end_layout \begin_layout Plain Layout \backslash renewcommand{ \backslash bibname}{References} \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset Flex TODO Note (inline) status open \begin_layout Plain Layout Check bib entry formatting & sort order \end_layout \end_inset \end_layout \begin_layout Standard \begin_inset CommandInset bibtex LatexCommand bibtex btprint "btPrintCited" bibfiles "refs" options "bibtotoc,unsrt" \end_inset \end_layout \end_body \end_document