MetaDrugSig

MetaDrugSig is a heterogeneity-aware transcriptomic drug signature database containing consensus gene expression signatures for 1,092 drugs generated from 6,593 independent RNA-seq comparisons across 3,683 GEO studies annotated using GEOMeta, an LLM-based metadata extraction pipeline. Drug signatures were computed using restricted maximum likelihood (REML) random-effects meta-analysis for drugs with ≥2 independent comparisons (Tier 1, n=556 drugs) or edgeR differential expression for single-study drugs (Tier 2, n=536 drugs). The database covers 63,474 genes and provides four gene-by-drug matrices: consensus log₂ fold changes (logFC), standard errors (SE), Benjamini-Hochberg adjusted p-values (FDR), and between-study variance (τ², Tier 1 only). Per-gene τ² enables the MDS signature framework — three filters (MDS-Broad, MDS-Robust, MDS-Context) of increasing stringency for heterogeneity-aware gene selection. A comparison metadata file provides tissue category (Uberon ontology), dose, and treatment duration annotations for 6,593 comparisons. The complete database is described in Joseph DF et al. Database (Oxford), 2026