Robustness of regulatory colocalization evidence at major depression risk loci

Updated lightweight research archive — 3 November 2026 This release accompanies “Robustness of regulatory colocalization evidence at major depression risk loci.” It builds on the previous corrective archive and adds external molecular analyses using BrainSeq, INTERVAL, and TenK10K data. Archive contents The archive includes revised manuscripts in Word/PDF formats; main and supplementary figures in PDF/SVG formats; supplementary results tables S1–S32; analysis code; selected derived inputs; model outputs; source metadata; provenance records; and portable numerical verification and figure-generation scripts. The corrected baseline analysis includes 10,966 gene–locus–tissue tests. Under default priors, 105 tests involving 88 genes at 55 loci met PP.H4 ≥ 0.8; four tests remained positive across all six prior configurations. The updated analyses evaluate selected candidates in external molecular cohorts and provide additional multi-signal and cell-resolved evidence. External molecular-cohort support is distinguished from independent disease GWAS replication, disease-state expression, and clinical generalization. Lightweight archive scope Large signed linkage disequilibrium (LD) matrices, redundant baseline result collections, and duplicate PNG/TIFF figures are omitted. The exclusions are documented in OMITTED_FULL_PACKAGE_FILES.csv. This archive supports selected numerical checks and figure reconstruction. Full SuSiE model refitting requires the matching signed LD matrices and other dependencies described in README.md. The archive does not provide a fully automated reconstruction of all source data and LD inputs from a new machine. Missing evidence, non-assessable models, and unsupported results are retained and distinguished. The findings do not establish causal biological mechanisms, disease specificity, or therapeutic efficacy. Exact participant-level overlap between source studies is not fully quantified. Correction history Historical version 1 remains superseded. Inappropriate variant-count normalization in posterior combination and cross-gene QTL record selection affected the earlier analyses. The former 130-locus “shared mechanism” category was an operational