Probabilistic Mechanism Graphs for GWAS Gene Prioritization - External Validations

This release provides a complete, self-contained reproduction package for "Mechanism‑First Causal Graphs for Noncoding GWAS". It contains the scripts, processed result files, source code, and a step‑by‑step guide needed to reproduce the paper’s quantitative claims from processed data. The package highlights the mechanism‑graph implementation (noisy‑OR aggregation + correlation corrections), calibration and benchmark artifacts, and concise reproduction instructions. Contents Reproduction scripts: reproduce_paper_claims.py, helper scripts Processed results: data/processed/* (locus summaries, calibration metrics, benchmark tables) Source snapshot: mechanism-graph-source-2026-01-21.zip Reproduction guide: DATA_REPRODUCTION_GUIDE.md, REPRODUCE.md Results summary: reproduction_results_summary.json License & citation: LICENSE, CITATION.cff Key reproduced results (included in package) Final calibration: ECE = 0.012 (per‑module ECE < 0.05) Benchmark recall@20: 0.76 [0.71–0.81] vs L2G 0.58 CRISPR benchmark: AUPRC = 0.71 [0.67–0.75] eQTL replication: 96.8% (exceeds claimed ≥78%) Quick reproduction Create environment: conda env create -f [environment.yml](http://_vscodecontentref_/6) && conda activate mechanism-gwas Run reproduction: python scripts/reproduce_paper_claims.py See full instructions: DATA_REPRODUCTION_GUIDE.md Notes Large raw/external data (>79 GB) are NOT included; processed inputs required for rapid reproduction are included.