from Eq. (1) to range between −4.5 and +2.5%. Although these differences were statistically significant in 11/18 simulation scenarios (two-tailed t-test, p-value <0.05/18), we found their sign to be inconsistent between ancestries. In fact, predicted RA from Eq. (1) slightly underestimates the observed RA in SAS and AFR ancestries but yields a small overestimation in EAS ancestry. Consequently, we found on average over all non-EUR ancestries, that Eq. (1) produces unbiased predictions, i.e. not statistically different from the observed RA (two-tailed t-test, p-value = 0.46). Similarly, we found our heuristic approach based on candidate causal variants to yield unbiased predictions of the RA in 7/18 simulated scenarios. More specifically, differences across scenarios between the mean observed RA and that predicted from Eq. (2) ranged between −5.3 and +5.2%. Note that this range is larger than when using information on causal variants. On average across simulation scenarios, we found that the approach assuming GWS SNPs to be the causal variants (referred to RApred3 in Fig. 1) strongly overestimated the RA in non-European ancestries. The average overestimation relative to the observed RA ranged from +4.3% (i.e. (96.3%−92.3%)/92.3% × 100%)) in SAS up to +103.0% in AFR ancestry. This result suggests that population
