We used data from 9 diverse ancestry studies to assess PGS transferability (Table S7A) using PGS derived from the clumping and thresholding approach. The PGSs were derived from the multi-ancestry and the European ancestry meta-analysis, excluding 23andMe (Neffective=739,180 and 576,327, respectively). In the diverse ancestry studies, the rl2, by the PGS based on the European ancestry training data ranged from ~0.6%–4.5%. The rl2 values for prediction into European ancestry (excluding 23andMe) were 3.9% (SE 0.2%) using PT = 0.05 (Figure 5; Table S7B). Values were lowest in studies with participants of African descent, and in the largest African ancestry study, the MVP, the PGS was not associated with MD rl2=0.0018). Results using the multi-ancestry summary statistics showed only minor and non-significant differences from European-only PGS GWAS-trained scores in all ancestry groups.
