Mirror Statistics-Based Gene Set Enrichment Analysis for Correlated Gene Expression Data

Gene set enrichment analysis (GSEA) is one of the most widely used approaches for identifying predefined gene sets enriched for differential expression between two experimental conditions. However, standard GSEA often fails to account for correlations among genes, as gene-level statistics are typically computed independently. In contrast, regularization methods such as the lasso and elastic-net jointly estimate regression coefficients in high-dimensional settings and have therefore been widely used to identify outcome-associated genes in correlated gene expression data. More recently, mirror statistics based on lasso estimates have been proposed for variable selection with false discovery rate control in high-dimensional models. In this study, we extend the mirror statistics framework to GSEA by incorporating various shrinkage estimators, including the lasso, elastic-net, debiased lasso, and ridge regression. Through extensive benchmark analyses, we demonstrate that mirror statistics based on ridge regression coefficients outperform competing approaches in terms of error control and detection power, particularly when gene expression data exhibit strong correlations. We further apply the proposed method to gene expression data from The Cancer Genome Atlas Breast Invasive Carcinoma cohort and identify biologically relevant cancer pathways that are not detected by conventional GSEA.