Gha-Hyun Jeffrey Kim
The 8q24 region is a major prostate cancer susceptibility locus, but variant effects may differ across ancestry-associated haplotype contexts. We analyzed whole genome sequencing and electronic health record data from 166,155 male participants in the NIH All of Us Research Program (10,166 cases and 155,989 controls). A random forest classifier trained to distinguish self-reported Black and White participants using 100 ancestry-informative variants generated an African regional ancestry proxy at 8q24 with 98.7% nested cross-validation accuracy. The primary scan tested 5197 variants and identified 35 genotype-by-ancestry-proxy interactions passing Bonferroni correction. A secondary SNV-restricted conditional analysis of 24 significant SNVs identified five independent signals. Four showed increased risk in the African-proxy group, led by rs116845582 (odds ratio 2.21 versus 0.94, interaction P = 5.2 × 10⁻¹¹), whereas rs7841060 showed the opposite interaction pattern (0.91 versus 1.28, interaction P = 1.8 × 10⁻⁹). All five retained their direction and significance when proxy probability was modeled continuously, across probability cutoffs, with 50 to 200 markers, and after correlation pruning. Regional ancestry-proxy context at 8q24 was associated with prostate cancer variant effects in both directions.