Reference distributions for the full PGS Catalog on the 1000 Genomes panel

Frozen scoring manifest and reference distributions for all 5,385 PGS Catalog models scored against all 3,202 individuals of the NYGC 1000 Genomes Project high-coverage release, produced by the memory-bounded GPU scorer described in the accompanying paper. The bundle contains the frozen scoring manifest, per-locus effect-allele frequencies in effect-allele orientation, PCA loadings for projection, the panel sample and target tables, and the raw, pooled-standardized, PC-adjusted and per-sample-coverage matrices, together with the build parameters, timings and thresholds. Reference means and standard deviations are computed over the 2,590 unrelated samples; the 612 trio children are scored but held out of the reference set. Median model coverage is 0.943, and 5,139 of 5,385 models pass the 0.75 coverage threshold; the remainder are flagged as not deployable rather than dropped. This release must not be applied to variant-only per-sample VCFs without a callable-region definition, because a locus absent from a sample's file is then ambiguous between hom-ref and uncalled, and the frozen imputation would be applied to genotypes that were in fact observed. Code: https://github.com/stormliucong/agent_optimized_pgs_calculation