Direction-Aware Two-Stage Computational Framework for Drug Repurposing in Immune Diseases: Knowledge Graph-Based Screening and Multi-Evidence Integration

Reproducibility archive for the study "A direction-aware two-stage computational framework for immune drug repositioning: knowledge-graph-derived screening and multi-evidence dossiers." This archive contains the knowledge graph, screening pipeline, scored candidates, and multi-evidence dossiers underlying the paper. Stage 1 (screening) enumerates 45 genes that complete a Drug—TARGETS→Gene—ASSOCIATED_WITH→Disease chain in the HIOCS v2.0 immune ontology (snapshot v3.28; 6,687 nodes / 18,024 edges) and ranks them by a four-component gene-level RepoScore*. Stage 2 (refinement) applies a novelty filter and evaluates four underexplored gene×disease hypotheses (CTLA4→Hashimoto thyroiditis, CD40→Crohn disease, CD40→Kawasaki disease, CD2→Graves disease) across six evidence axes, including risk-allele eQTL directionality. Contents (42 files): the immune knowledge graph (JSON), the screening pipeline and environment specification, 17 result CSVs (scorecard, gold sets, dossiers, baseline, eQTL, audits, disease mapping), 12 provenance/statistics JSONs, four figures, the manuscript source, a README, and an integrity bundle (SHA-256 manifest + verification script). All external data were retrieved as a fixed snapshot on 2026-07-23: Open Targets Platform (release 26.06 / API 26.6.3), GTEx v8 (GENCODE v26), GWAS Catalog, ChEMBL, RCSB PDB / AlphaFold DB, and ClinicalTrials.gov / Drugs@FDA / openFDA. File integrity is verifiable via SHA256SUMS.txt (shasum -a 256 -c) or verify_manifest.py. This is a computational/methodological study; it does not include wet-lab, in-vivo, or clinical efficacy data. The four hypotheses are testable predictions for experimental validation, not efficacy claims. AI (LLM-based) research-assistant tools were used for analysis, drafting, and translation; all values and conclusions were verified by the authors.