SexDiffKG v3: A Sex-Differential Drug Safety Knowledge Graph from 14.5 Million FDA Adverse Event Reports

SexDiffKG is a sex-differential drug safety knowledge graph constructed from 14,536,008 FDA Adverse Event Reporting System (FAERS) reports spanning 2004–2024, integrated with molecular target data from ChEMBL 36, protein interaction networks from STRING v12.0, and biological pathway annotations from KEGG and UniProt. Key statistics: 127,063 nodes (6 entity types: drugs, adverse events, genes, pathways, proteins, sex-differential expression markers) 5,839,717 edges (6 relation types) 49,026 strong sex-differential signals (|ln(ROR ratio)| > 1.0, ≥10 reports per sex) 58.5% female-biased signals 429 gene targets with sex-bias profiles Embedding results: DistMult v3 (200d, 100 epochs): MRR 0.048, AMRI 0.981, Hits@10 8.85% RotatE v3 (200d complex, 25 epochs): Near-random (AMRI 0.003) — validates DistMult as primary model Contents: Full study document, ISMB 2026 abstract, 10 supplementary tables, 5 supplementary figures (PNG + PDF), analysis data, embedding metrics, pipeline scripts, molecular audit report, and KG node/entity mappings. Built on NVIDIA DGX Spark (Grace Blackwell GB10, ARM64, 128GB unified memory).