For each individual, an estimated PRS was calculated as the sum of the log(OR) (i.e., the PRS “weights”) multiplied by their genotype for all independent and significant variants at a given threshold. The PRSs were constructed for testing samples with variants and weights each selected from European or African ancestry GWASs or a fixed-effects meta-analysis of both combined. Additional multi-ancestry PRS approaches7,10 were also explored for admixed individuals. Accuracy was measured by Pearson’s correlation (r) between the true genetic liability and estimated PRS within each population. Across simulation trials, averages and 95% confidence intervals (CIs) for r were calculated following a Fisher z-transformation for approximate normality.14 The statistical significance of differences in accuracy between PRS approaches was assessed within ancestry groups, defined by global genome-wide European ancestry proportions, with a Z test (also following Fisher transformation). Specifically, within each simulation trial the z-statistic, measuring the difference between two PRS approaches, was computed, and a two-sided p value was obtained; results were summarized across trials by taking the median p value. While using r as a measure of accuracy has the
