We used PRS-CSx, a recently developed method designed for cross-ethnic polygenic prediction that showed better performance than other methods in simulation studies and real data analysis [23]. The posterior effect size of each variant was estimated via a Bayesian regression framework using continuous shrinkage priors. African and European samples from the 1000 Genomes Project were used as the LD reference panels. PRS-CSx can estimate posterior effect sizes of AA only, EA only, and meta-analysis of EA-PAU and AA-AUD. The authors of PRS-CSx recommend using estimated AA- and EA-only posterior effect sizes, then testing different linear combinations of them with different weights in a validating dataset, and choosing the one with the best performance for testing in independent datasets [23]. If the validation dataset and independent datasets are similar, e.g., having similar LD patterns and allele frequencies, this method will have more power. However, if they are different, then the weights estimated from the validating dataset will be biased toward that dataset and different from the independent datasets, resulting in loss of power. As we noted earlier, AA is a very
