We next applied our method to predict the fraction of RA attributable to MAF and LD differences between ancestries (Fig. 3). We used WGS data from the 1KGP as a reference panel for LD and allele frequency calculations. For traits with significantly reduced RA, we define the loss of accuracy (LOA) of their corresponding PGS as LOA = (1 − RA) × 100%. We also define the proportion of LOA explained by LD and MAF as the ratio between the predicted LOA over the observed LOA (Supplementary Fig. 12). The derivation of standard errors of this proportion is given in Supplementary Note 7. Our method predicts a different contribution of LD and MAF across traits and ancestries (Supplementary Table 2, Fig. 3). For example, we predict that MAF and LD differences between EUR and SAS ancestries explain ~24% (S.E. 2%) of the LOA of the height PGS, while we predict those factors to explain ~38% and ~72% of the LOA in EAS and AFR ancestry, respectively. For T2D and HTN, respectively, we found that ~82% (S.E. 14%) and ~86% (S.E.
