To help answer these questions, here we systematically and empirically explore the relationship between PRS performance and ancestry within African, European, and admixed ancestry populations through simulations. We highlight PRS-building approaches that will achieve unbiased estimates across global populations and admixed individuals with future recruitment and representation of non-European ancestry individuals in GWASs. We also investigate reasons for loss of PRS accuracy and attribute this to population differences in LD tagging of causal variants by lead GWAS variants, as well as allele frequency biases potentially due to genetic drift undergone by European ancestry populations. Finally, we confirm our simulation findings by application to data on hemoglobin A1c (HbA1c) levels, asthma, and prostate cancer in individuals of European and individuals of African ancestry from the UK Biobank.
