FedEdgeR: federated and privacy-preserving edgeR for differential gene expression analysis
Motivation: Multi-center RNA-seq studies improve statistical power, but privacy regulations restrict patient-level data sharing. Meta-analysis methods avoid this restriction but lose power, especially under per-site imbalance. Federated learning allows sites to share only summary statistics. Among the three dominant differential expression (DE) frameworks, Flimma federates limma-voom and…
Motivation for the development of Federated and Privacy-Preserving edgeR (FedEdgeR) stems from the need to improve statistical power in multi-center RNA-seq studies while respecting privacy regulations that restrict patient-level data sharing. Meta-analysis methods, which pool data from multiple studies, can bypass these restrictions but often sacrifice power, particularly when faced with imbalances in sample sizes across sites.
Federated learning offers a solution by enabling sites to collaborate without sharing raw data, instead exchanging summary statistics.
Among the three leading differential expression (DE) frameworks, Flimma federates limma-voom and FedPyDESeq2 federates DESeq2. However, edgeR, a method frequently favored for its performance in small-sample and high-variability designs, has remained without a federated counterpart. This is due to edgeR's reliance on iteratively reweighted least squares (IRLS) and Cox-Reid dispersion estimation, which pose challenges for secure federation compared to limma-voom's more straightforward single-pass fitting approach.
FedEdgeR, introduced in this work, addresses this gap by presenting a federated implementation of edgeR that employs secure multi-party computation (SMPC) to ensure data privacy. FedEdgeR incorporates IRLS GLM fitting, three-level dispersion estimation, and the likelihood-ratio test, allowing it to handle the unique aspects of edgeR's methodology.
To assess its performance, FedEdgeR was evaluated on four diverse RNA-seq datasets, including two tumor/normal cohorts, a multi-cohort anti-PD-1 immunotherapy study, and a 6-sample paired stress-test.
The results demonstrate that FedEdgeR delivers comparable performance to pooled edgeR across various metrics. Specifically, it achieves Pearson's correlation coefficients (r) greater than or equal to 0.99999 on -log10 p-values, maintains 100% complete overlap of the top-100 DE genes, and attains an F1 score of 1.000 at a false discovery rate (FDR) of 0.05.
Crucially, FedEdgeR consistently outperforms Fisher's, Stouffer's, random-effects, and RankProd meta-analysis methods when tested on every dataset, showcasing its effectiveness in maintaining both privacy and analytical robustness in federated settings.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
