For biologists analyzing RNA sequencing data, the choice between edgeR and DESeq2, the two most widely used tools for differential gene expression analysis, has often been a matter of personal preference or institutional habit. A new study published in PLOS ONE suggests that the choice matters more than many researchers assume.
The study, led by Mostafa Rezapour, evaluated current implementations of edgeR (v4.4.2) and DESeq2 (v1.46.0) across a diverse collection of real-world bulk RNA-seq datasets spanning viral infection, bacterial infection, and fibrotic lung disease in both human and nonhuman primate systems.
What was measured
The evaluation framework covered four dimensions: sensitivity to sample size and robustness to outlier perturbation; classification performance of genes uniquely identified by each tool; pathway-level concordance of enriched biological processes; and cross-study generalizability across independent datasets.
For the cross-study analysis, the study used four independent SARS-CoV-2 datasets to test whether gene sets identified by each tool in one dataset could separate disease from control samples in held-out datasets.
The results
On sensitivity and robustness, both tools performed similarly. Jaccard similarity between differentially expressed gene (DEG) sets from perturbed and original data decreased as more outliers were added, at comparable rates for both methods.
The divergence emerged on classification performance and generalizability. Classification models trained on tool-specific genes showed that edgeR achieved higher F1 scores in 9 of 13 contrasts and more frequently reached perfect or near-perfect precision. Dolan-More performance profiles indicated edgeR maintained performance closer to optimal across a greater proportion of datasets.
In the cross-study validation, the difference was pronounced. Gene sets uniquely identified by edgeR yielded higher AUC, precision, and recall in classifying samples from held-out SARS-CoV-2 datasets, a pattern consistent across folds, with some test cases achieving perfect separation using edgeR-specific genes. DESeq2-specific genes showed lower and more variable performance across studies.
However, DESeq2 identified more DEGs overall, even under stringent significance thresholds. The tradeoff, the authors conclude, is between discovery sensitivity and reproducibility.
What this means for researchers
“A key question in differential expression analysis is not only which tool identifies more genes, but which tool identifies gene sets that are more stable, biologically interpretable, and transferable across studies,” Rezapour writes.
The findings suggest that for biomarker discovery, clinical transcriptomics, or any application where cross-study reproducibility is critical, edgeR may be the more reliable choice despite identifying fewer candidate genes. For exploratory studies where maximizing discovery is the priority, DESeq2’s broader gene detection may be preferable, but the results should be interpreted with awareness that some of those genes may not replicate in independent datasets.
The study also tested the common practice of intersecting results from both tools as a validation strategy. The analysis found this approach reduces sensitivity without necessarily improving robustness, since both tools share the same core statistical foundation, negative binomial generalized linear models, and their disagreements tend to cluster around borderline signals.
“Avoid intersecting results from both tools and calling it validation,” the authors advise. “Use diagnostics and validate the biology.”
Sources
Rezapour M. “Tool Choice Matters: Evaluating edgeR vs. DESeq2 for Sensitivity, Robustness, and Cross-Study Performance.” PLOS ONE (2026). DOI: 10.1371/journal.pone.0353788