Evaluating bias in polygenic score associations due to imperfect control of population stratification

Principal component analysis is the workhorse correction for population stratification, but its behaviour in the regime that matters most — very large samples, many markers, diverse ancestry panels — is not what intuition suggests.

This project uses random matrix theory to characterize the limitations of PCA-based correction in large-scale genetic association studies, and quantifies the bias that residual ancestry stratification introduces into polygenic score associations and GWAS summary statistics.