To understand the relationship between ancestry and PRS accuracy, admixed individuals were stratified by their proportion of genome-wide European (CEU) ancestry: high (100% > CEU > 80%), intermediate (80% > CEU > 20%), and low (20% > CEU > 0%). PRS performance decreased with decreasing European ancestry (Figure 1). Average accuracy (Pearson’s correlation) for the high, intermediate, and low European ancestry groups was 0.73 (95% CI = [0.72, 0.74]), 0.61 (95% CI = [0.60, 0.62]), and 0.53 (95% CI = [0.51, 0.54]), respectively (Figure 1). In comparison to Europeans, the performance of the European-derived PRS was significantly lower in individuals with intermediate (20% decrease; p = 1.27 × 10−47), and low (32% decrease; p = 6.48 × 10−16) European ancestry, and with African-only ancestry (41% decrease, p = 8.00 × 10−155). There was no significant difference for individuals with high (5.3% decrease; p = 0.09) European ancestry. These trends remained consistent when varying the genetic architecture (Figure S1), specifically decreasing the number of causal variants (m) and varying the trait heritability (h2). Additionally, the relationship between ancestry and accuracy persisted with the inclusion of variants at lower p value thresholds (Figure S2).
