What problem does it solve? Standard gene-set enrichment methods over-detect hits: Fisher's exact discards rank information, ROC-AUC calls almost any large set enriched, and highly multifunctional genes (TP53, TNF) make sets look significant regardless of the underlying biology. This Skill applies the stringent PR-AUC method from Ballouz et al. (2017) plus multifunctionality correction so enrichment results reflect the user's specific signal rather than generic gene properties. ## Core Features & Use Cases - PR-AUC enrichment (pr_enrichment): Takes a scored/ranked gene list and computes per-set PR-AUC with permutation p-values, BH q-values, and multifunctionality-corrected p-values against GO BP or any GMT library (KEGG, Reactome, MSigDB). - Over-representation analysis (ora): Hypergeometric/Fisher's exact test for hit lists, with a parallel MF-baseline run using the top-n most multifunctional genes to flag non-specific hits. - Full provenance: Every run writes a .meta.json sidecar recording input sha256, GAF date, OBO version, background set hash, permutation count, and seed. - Use Case: Given a TSV of differential expression scores, run pr_enrichment to rank GO biological process terms by PR-AUC and immediately see which hits survive multifunctionality correction. ## Quick Start Ask the assistant to run PR-AUC gene-set enrichment on your scored gene list TSV against the default GO biological process library with multifunctionality correction enabled.
