Let us consider an admixed genome which descends with proportion p from population A and with proportion 1−p from population B. We can compute a PS for this individual adopting summary statistics obtained from a GWAS performed in population A, PSA. We define a model to simulate such admixed genome by combining an unbiased PS with a biased and poorly predictive PS with a proportion p in one single individual. As expected in case of poor PSA transferability, the proportion of ancestry A (p) is positively correlated with predictability of PSA for the trait of interest (Supplementary Fig. 1, yellow line).
