Using a linear mixture approach developed by Márquez-Luna et al.,10 we combined two PRSs estimated in each of our global training populationsPRS=α1PRSEUR+α2PRSAFRwhere individual PRSs were constructed using significant and independent variants (p < 0.01 and r2 < 0.2 in a 1Mb window) and effect sizes from a GWAS in that training population. For simulations, mixing weights (α1 and α2) were estimated in an independent African ancestry testing population, and as validation, accuracy was assessed in our simulated admixed ancestry individuals.f
