variancePartition: interpreting drivers of variation in complex gene expression studies.

Hoffman GE, Schadt EE. · 2016

Read variancePartition: interpreting drivers of variation in complex gene expression studies. by Hoffman GE, Schadt EE. (2016) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5123296.txt.

Introduction

Gabriel E. Hoffman and Eric E. Schadt presented a software tool in BMC Bioinformatics in 2016 for sorting out what drives gene activity data. Big studies now mix many influences at once, from disease state and sex to tissue type and lab technique, and untangling them is needed before biology can be read.

Their program, variancePartition, fits a statistical model for each gene and reports the share of its variation tied to each factor, then flags genes that break the genome-wide pattern. On four large data sets it recovered clear, repeatable patterns of both biological and technical variation. The tool is free through the Bioconductor project and also suits other large-scale genomics assays, the authors note.

Catalog record

CREATORHoffman GE, Schadt EE.
YEAR2016
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC5123296

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

variancePartition: interpreting drivers of variation in complex gene expression studies By Gabriel E Hoffman, Eric E Schadt Published in: BMC Bioinformatics DOI: 10.1186/s12859-016-1323-z ABSTRACT Abstract Background As large-scale studies of gene expression with multiple sources of biological and technical variation become widely adopted, characterizing these drivers of variation becomes essential to understanding disease biology and regulatory genetics. Results We describe a statistical and visualization framework, variancePartition, to prioritize drivers of variation based on a genome-wide summary, and identify genes that deviate from the genome-wide trend. Using a linear mixed model, variancePartition quantifies variation in each expression trait attributable to differences in disease status, sex, cell or tissue type, ancestry, genetic background, experimental stimulus, or technical variables. Analysis of four large-scale transcriptome profiling datasets illustrates that variancePartition recovers striking patterns of biological and technical variation that are reproducible across multiple datasets. Conclusions Our open source software, variancePartition, enables rapid interpretation of complex gene expression studies as well as other high-throughput genomics assays. variancePartition is available from Bioconductor: http://bioconductor.org/packages/variancePartition. Electronic supplementary material The online version of this article (doi:10.1186/s12859-016-1323-z) contains supplementary material, which is available to authorized users. Keywords: Transcriptome profiling, RNA-seq, Linear mixed model BACKGROUND High-throughput genomics assays have revolutionized our understanding of the molecular etiology of human disease, molecular biology of cell lineage and genetic regulation of gene expression. Transcriptome profiling in particular has been widely applied to detect variation in transcript levels attributable to differences in disease state, cell type or regulatory genetics. As transcriptome profiling studies have expanded in size and scope, they have grown increasingly complex and consider multiple sources of biological and technical variation. Recent studies have simultaneously considered multiple dimensions of variation to understand the impact of cell type [1], tissue type [2], brain region [3], experimental stimuli [4], time duration following stimulus [5] or ancestry [1, 4, 6] on the genetic regulation of gene expression. More studies are including a disease axis, for example to characterize the role of regulatory genetics on late onset Alzheimer’s disease in multiple brain regions [7]. The fundamental challenge in the analysis of complex datasets is to quantify and interpret the contribution of multiple sources of variation. Indeed the most pressing questions concern the relationship between these sources of variation. How does cell or tissue type affect the genetic regulation of gene expression, and does it vary by ancestry [1, 2]? What is the relative contribution of experimental stimulus versus regulatory genetics to variation in gene expression [5]? Is technical variability of RNA-seq low enough to study regulatory genetics and disease biology, and what are the major drivers of this technical variability [2, 8, 9]? A rich understanding of complex datasets requires answering these questions with both a genome-wide summary and gene-level resolution. Standard computational workflows employ principal components analysis [10] and hierarchical clustering [11] to summarize genome-wide expression patterns, and differential expression [12–16] to perform gene-level analyses. Recently, statistical methods that decompose variation in gene expression into the variance attributable to multiple variables in the experimental design have yielded valuable insight into the biological and technical components driving expression variation [8, 17–22]. Moreover, linear mixed models have been widely used in the analysis and interpretation of genome-wide association studies [23–28]. The linear mixed model is uniquely suited to interpreting drivers of variation in complex gene expression studies. Yet the lack of a convenient workflow and scalable implementation for analysis and visualization have prevented wider application of this rich statistical framework. Applying this analysis framework to gene expression data currently requires particular expertise in regression modeling, computational statistics, the R programming language and data visualization. Even then, the time required to implement the analysis is often prohibitive. As gene expression datasets become more complex, the analysis and interpretation of the data is becoming the rate-limiting step. We have developed the variancePartition software and workflow to facilitate rapid analysis and improve interpretation of complex gene expression datasets. The software and workflow enables any analyst to perform a sophisticated analysis and visualize the results in hours using a few lines of R code. variancePartition leverages the power of the linear mixed model [29–31] to jointly quantify the contribution of multiple sources of variation in high-throughput genomics studies. In applications to transcriptome profiling, variancePartition fits a linear mixed model for each gene and partitions the total variance into the fraction attributable to each aspect of the study design, plus the residual variation. Because it is built on the first principles of the linear mixed model, variancePartition has well characterized theoretical properties [29–31] and accurately estimates the variance fractions even for complex experimental designs where the standard ANOVA method is either inaccurate or not applicable. Moreover, variancePartition gives strong interpretations about the drivers of expression variation, and we demonstrate that these findings are reproducible across multiple datasets. Here we apply variancePartition to four well-characterized gene expression studies to demonstrate how the workflow facilitates interpretation of drivers of expression variation in complex study designs with multiple dimensions of variation. We illustrate how variancePartition enables rapid interpretation of the drivers of expression variation in these complex datasets. IMPLEMENTATION Overview of the software The variancePartition R package implements a computational workflow (Fig. 1) that is complementary to standard analyses and provides particular insight into datasets with multiple dimensions of variation. variancePartition provides a user-friendly, parallelized interface for genome-wide analysis and publication quality visualizations to examine the results. Because the variance fractions are simple to describe and interpret, variancePartition can give particular insight into how each dimension of variation contributes to transcriptional variability. A typical variancePartition analysis comprises: 1) fitting a linear mixed model to quantify the contribution of each dimension of variation to each gene, 2) visualizing the results, and 3) integrating additional data about each gene to interpret drivers of this variation. The variancePartition workflow requires only a few lines of R code for pre-processing, analysis and visualization and this enables rapid interpretation of complex datasets. [Figure: Analysis workflow of gene expression data and meta-data. Standard analysis consists of interpreting gene expression data with respect to variables in the metadata using genome-wide analysis such as a principal components analysis and b hierarchical clustering, and gene-level analysis such as c differential expression. The variancePartition workflow uses a rich statistical framework in the form of a linear mixed model and produces gene-level results and a genome-wide summary to simultaneously interpret gene expression data in the context of multiple variables in the metadata. The workflow produces d gene-level results quantifying the contribution of each metadata variable to the variation in expression of each gene, and e a violin plot to summarize the genome-wide trend and rank the total contribution of each variable. f The gene-level results can be used to identify genes that show high expression variation across individuals (i.e. gene385) or tissue (i.e. gene644). Furthermore, variancePartition facilitates examination of specific genes, and integrating external data enables further interpretation of the drivers of expression variation] The variancePartition software is implemented in R and is optimized for genome-wide analysis of large-scale transcriptome profiling datasets. variancePartition uses the packages lme4 [29] foreach [32], iterators [33] and doParallel [34] to efficiently fit a linear mixed model for each gene in parallel on a multicore machine with a small memory footprint. The

Read the complete text