Statistical Methods for Identifying Risk Variants in Multi-Ancestry Genome-Wide Association Studies

Genome-wide association studies (GWAS) have successfully identified genetic variants associated with complex diseases. However, they face significant challenges, including high dimensionality, complex linkage disequilibrium, and ancestral heterogeneity, which compromise statistical power and reproducibility. While standard false discovery rate (FDR)control methods offer advantages over conservative family-wise error rate approaches, they often suffer from inflated FDR in the presence of complex variant correlations. The knockoff framework has emerged as a powerful alternative providing rigorous FDR control under arbitrary dependency structures, yet key limitations remain in multi-ancestry applications.This project proposes three novel knockoff methods for multi-ancestry GWAS. First, because conventional knockoffs assume genetic homogeneity and suffer confounding from population structure, we propose a semiparametric linear mixed model-based knockoff construction that accounts for ancestral heterogeneity to achieve robust FDR control. Second, to overcome the heavy computational burden of individual-level data in large-scale genetics, we develop a knockoff-augmented meta-analysis framework leveraging only summary statistics for multi-ancestry meta-analysis with rigorous FDR control. Third, because European-derived GWAS findings often fail to generalize, we propose a multi-knockoff method to identify risk variants demonstrating consistent associations across diverse populations. Preliminary results show that our proposed methods consistently outperform existing competing approaches in most scenarios.