ssGSEA3.0: Feature-Level Bootstrap Resampling for Robust Assessment of Statistical Significance

Abstract

Motivation Single-sample Gene/Post-Translational Modification Set Enrichment Analysis (ssGSEA/PTM-SEA) methods assess pathway or signature activity per sample using permutation-based significance testing. ssGSEA2.0's weighted scoring (ssGSEA2.0-orig) uses a correlation-vector reordering that inflates false positives at nonzero weight; directly fixing this (ssGSEA2.0-fixed) restores null-distribution calibration but sacrifices detection power at nonzero weight. Results ssGSEA3.0 replaces feature-label (gene- or site-level) permutation with feature-level bootstrap resampling, which remains calibrated across weighting schemes. On published lung adenocarcinoma and phosphoproteomic perturbation datasets, ssGSEA3.0 shows greater detection power than ssGSEA2.0-fixed, with an additional gain for directional (PTM-SEA) scoring. Previously published findings remain robust across methods, including at nonzero weight. Bootstrapping adds a ~1.5- to 2-fold computational cost. Availability and Implementation ssGSEA3.0 is implemented as an open-source R package (github.com/broadinstitute/ssGSEA3.0, DOI: 10.5281/zenodo.23734262) and includes the worked examples on biological data presented here.