Skip to content

edgeR

Empirical Analysis of Digital Gene Expression Data in R — the foundational Bioconductor package for differential expression analysis of sequence count data using negative binomial distributions, including empirical Bayes estimation, exact tests, GLMs, quasi-likelihood methods, and gene-set enrichment; supports RNA-seq, ChIP-seq, ATAC-seq, Bisulfite-seq, SAGE, CAGE, metabolomics, and proteomics spectral counts at gene or isoform level.

Talk at EuroBioC2025

  • Speaker: Lizhong Chen (WEHI)
  • Day / session: Day 1 (Wed Sep 17, 2025) — afternoon short-talk session, after the Holmes keynote
  • Talk title: "edgeR v4 functionality expansion"
  • Slides (Zenodo DOI): to verify after publish
  • Video: not recorded — short talks at EuroBioC2025 were not part of the plenary YouTube playlist
  • Abstract / schedule: eurobioc2025 schedule

What it does

edgeR fits negative-binomial GLMs to count data and tests for differential abundance using exact tests, likelihood-ratio tests, or quasi-likelihood F-tests, with empirical-Bayes shrinkage of dispersions. It is technology-agnostic across count-based assays — RNA-seq, ChIP/ATAC-seq, BS-seq, SAGE/CAGE, spectral-count proteomics — and operates at gene or isoform resolution.

How it works

Core idea. Counts are modeled as negative binomial with a gene-specific dispersion. Dispersions are estimated by Cox-Reid profile-adjusted likelihood and shrunk toward a common or trended value with empirical Bayes, so that low-count genes borrow strength from the dataset as a whole. The standard differential-abundance test is the quasi-likelihood F-test (glmQLFTest), which adds a second-level QL dispersion to account for uncertainty in the gene-wise dispersion estimate — giving more reliable type-I error control on small replicate sets than the older likelihood-ratio test.

Inputs / outputs. Input: an integer count matrix (genes × samples) and a design matrix; wrapped as a DGEList. Output: a DGELRT / DGEGLM result object with log-fold-changes, p-values, FDR; topTags() ranks the gene table.

Key innovation (and what v4 added). The original innovation was the NB-with-empirical-Bayes-dispersion framework — the model that effectively defined RNA-seq DE testing alongside DESeq/DESeq2 and limma-voom. The v4 release (2025) is described by the authors as extending edgeR with expanded functionality and improved handling of small counts and larger datasets — relevant for single-cell-scale matrices, which is what the EuroBioC2025 v4-expansion talk foregrounds.

Parameters worth knowing. filterByExpr() — filter low-count genes before dispersion estimation. normLibSizes() (formerly calcNormFactors) — TMM normalization by default. estimateDisp(..., robust = TRUE) — robust empirical-Bayes shrinkage that down-weights outlier genes. Choice of glmQLFit + glmQLFTest (recommended) versus the older glmFit + glmLRT pipeline.

Canonical example. The introductory vignette walks through a standard RNA-seq design: build a DGEList, filterByExpr, normLibSizes (TMM), estimateDisp against a model matrix, glmQLFit + glmQLFTest for the contrast of interest, then topTags for the ranked DE table.

Where it fits in the corpus

  • AACR 2026: no current dossier as a standalone tool, but edgeR is the substrate underlying many AACR transcriptomic analyses
  • DESpace (entry) — builds directly on edgeR's NB modeling for spatial cluster testing
  • GBCC 2025: queued
  • Nextflow Summit 2026: not mentioned directly; nf-core RNA-seq pipelines commonly hand off to edgeR downstream

Notes

edgeR is the substrate for many other Bioconductor packages; the v4 expansion broadens single-cell and sparse-count handling, which is what the WEHI talk is foregrounding.