Differential meta-analysis of RNA-seq data from multiple studies

Abstract

BackgroundHigh-throughput sequencing is now regularly used for studies of the transcriptome (RNA-seq), particularly for comparisons among experimental conditions. For the time being, a limited number of biological replicates are typically considered in such experiments, leading to low detection power for differential expression. As their cost continues to decrease, it is likely that additional follow-up studies will be conducted to re-address the same biological question.ResultsWe demonstrate how p-value combination techniques previously used for microarray meta-analyses can be used for the differential analysis of RNA-seq data from multiple related studies. These techniques are compared to a negative binomial generalized linear model (GLM) including a fixed study effect on simulated data and real data on human melanoma cell lines. The GLM with fixed study effect performed well for low inter-study variation and small numbers of studies, but was outperformed by the meta-analysis methods for moderate to large inter-study variability and larger numbers of studies.ConclusionsThe p-value combination techniques illustrated here are a valuable tool to perform differential meta-analyses of RNA-seq data by appropriately accounting for biological and technical variability within studies as well as additional study-specific effects. An R package is available on the CRAN (http://cran.r-project.org/web/packages/metaRNASeq).

Unmasking Transcriptional Heterogeneity in Senescent Cells

“…For RS, the only group with more than one dataset, we performed a meta-analysis combining three different methods: we used (1) a negative binomial-generalized linear model (NB-GLM), pooling all the samples in all the datasets for one particular senescence-inducing stimulus and comparing them to proliferating counterparts, and within the model, we included a covariate that accounted for inter-laboratory and inter-strain differences (see STAR Methods for details); (2) an analysis of each individual dataset and subsequent combination of the p values using the Fisher method; and (3) an inverse-normal p value combination technique in which each dataset was weighted according to the number of replicates [17, 18]. We set a stringent threshold of nominal p ≤ 0.01 to reduce the odds of false-positive results and retained only those genes that were differentially expressed by the three different methods.…”

Methods to increase reproducibility in differential gene expression via meta-analysis

“…For this reason, random-effects models are generally a more desirable method to use in gene-expression meta-analysis, where the real goal is to try to discover the background biological ‘true’ effect, rather than simply to synthesize the available data. One drawback of random-effects meta-analysis is that it is not appropriate for count-based data such as RNA-seq (as the count data are not normally distributed), as discussed elsewhere (27,28). However, microarrays are still the dominant genome-wide expression measurement assay: in 2015, 6569 new RNA assays were indexed by ArrayExpress and GEO, of which 2024 were sequencing assays and 4615 were array assays (presumably a very small number of studies had both).…”

Statistically robust methylation calling for whole-transcriptome bisulfite sequencing reveals distinct methylation patterns for mouse RNAs

“…Because secondary structures and nucleic acid binding proteins cause reproducible bisulfite deamination artifacts (Warnecke et al 2002), we could not use raw P-values for subsequent tests, as this would lead to significance for most of these artifacts. However, Fisher's method has shown robustness in a variety of contexts where the underlying assumptions are not strictly met (Derkach et al 2013;Rau et al 2014). We therefore summed the log-transformed www.genome.org…”