Supporting data and code for: Terazosin analog 7d shows low predicted α1-adrenergic binder-likelihood across subtypes and high predicted PGK1 binder-likelihood in a Parkinson's disease repurposing analysis

Complete supporting materials for a computational triage study of terazosin analogs in Parkinson's disease. Contains 200 Boltz-2 predictions (10 ligands × 4 targets [PGK1, ADRA1A, ADRA1B, ADRA1D] × 5 seed-labelled replicates), the analysis code that regenerates every reported statistic and figure, all 200 model input files, target sequences, an independently reproducible AutoDock Vina docking cross-check with full provenance, manuscript figures, the preprint, and verification manifests. All values are model predictions. Predicted binder-likelihood is not measured binding affinity, activation, functional antagonism, clinical selectivity, or evidence of safety. Nothing here is medical or prescribing guidance. Known provenance gaps are documented in README.md, including the absence of per-job raw model outputs, cached MSAs, and a complete environment lock for the Boltz-2 campaign. The docking cross-check was re-executed under a fixed random seed with all inputs, configurations, logs and outputs deposited. Code in code/ is licensed MIT; the manuscript, data, model inputs and figures are licensed CC BY 4.0.