As anticipated by our simulations, the main advantage of computing aspPS over simple PS lies in the possibility of combining multiple aspPS into a single, global, ancestry-informed PS. We therefore focused on samples for which at least two aspPS could be computed (i.e. individuals with East Asian and European ancestral components), thanks to the availability of European- and East Asian-derived trait-SNP associations, from UKBB and BBJ respectively. When adding together the two aspPS weighted for ancestry proportions, to generate a combined ancestry specific PS (casPS), we show that the phenotype predictivity outperforms both aspPS and at least one of the total PS (Fig. 5b, d purple line). Again, this is not achieved with incorrect local ancestry inference (Supplementary Fig. 6b, d). Different fitting sets were adopted: no macroscopic differences were evident when fitting the model parameters on either admixed samples alone or European samples only, see Supplementary Fig. 7. Lastly, we compared models including casPS with the traditional total PS, performing a Vuong’s closeness test to detect which one fared better when predicting traits in admixed samples. For BMI, our
