individual’s PS and, when multiple aspPS are combined in a casPS, the casPS predictivity outperforms both aspPS and at least one of the total PS scores (European or East Asian total PS, in our examples). Intuitively, and according to our simulation, casPS performance is enhanced by the lack of PS transferability across ancestries and if the Non-European PS performs at least as well as the European one on Non-European sample sets. Our results show that this is not the case, since PSs calculated using BBJ (East Asian) associations perform worse on East Asians than PSs calculated using UKBB associations, regardless of the adopted fitting (Fig. 5 and Supplementary Fig. 7). The difference in overall performance between the two PSs on the respective unadmixed individuals, possibly due to differences in statistical power between the underlying GWAS, might be the reason why casPS for height did not show a significant improvement over the total PS calculated using UKBB trait-SNP associations (Fig. 5b), although it did compared to total PS computed with weights from BBJ. In either cases, however, we maintain that aspPS and casPS are preferable measures than the simple PS, despite their comparable predictive performance, due to the fact that their
