We constructed PRSs from our simulated European datasets and applied them to independent simulated European, African, and admixed testing populations, assuming 1,000 true causal variants (m) and trait heritability (h2) of 0.5. On average, 1,552 (range = [1,134–1,920]) variants were selected for inclusion into the PRS at p value < 0.01 and LD r2 < 0.2 (Table 1). The average accuracy across replicates (50 simulations), measured by the correlation (r) between the true and inferred genetic risk, was much higher when applying the PRS to Europeans (r = 0.77; 95% CI = [0.76, 0.77]) than to Africans (r = 0.45; 95% CI = [0.44, 0.47]; Figure 1). This is in agreement with decreased performance seen in real data when applying a European-derived genetic risk score to an African population.2, 3, 4, 5Table 1Summary of PRS variants and causal tagging across simulationsGWAS populationTotal No. PRS variants (p < 0.01)No. causalNo. in LD with a causal variantr2 > 0.8r2 > 0.6r2 > 0.4r2 > 0.2EuropeanLD in Europeans1,552 (1,134–1,920)18 (10–26)27 (16–40)32 (22–44)39 (25–55)58 (38–80)LD in Africans20 (9–36)25 (16–42)34 (24–54)53 (35–70)AfricanLD in Europeans5,269 (4,462–6,071)28
