transformation proposed previously by Lee and colleagues31 (Methods). Note that using ancestry-specific prevalence from previous population studies32–35 did not change our results (Supplementary Fig. 9). We also assessed the predictive accuracy of PGS based upon sub-significant SNPs (Supplementary Note 6) and found, in individuals of non-European ancestries, that PGSs including SNPs selected at less stringent p-value thresholds did not systematically improve over using GWS SNPs only (Supplementary Fig. 10). This important observation is consistent with previous studies2,3,19,20 and emphasises that observed RA from p-value thresholding scoring methods can be seriously underestimated if based on sub-significant SNPs.
