As shown in Fig. 3, methodological choices in the construction of polygenic risk scores can cause dramatic differences in distributions of polygenic risk scores for worldwide populations (polygenic scores were constructed for all 1000Genomes participants, N = 2577, see Methods for additional details). In general, the inclusion of more variants caused greater dispersion of distributions for these 1000Genomes populations (i.e., comparing panel A with panel B; genome-wide significant variants to all variants). Lower r2 thresholds for clumping tended to make worldwide population distributions more similar (left to right, in both A and B), and use of particular populations for clumping also dramatically affected distributions, particularly when East and South Asian populations were used for clumping. For further demonstration of the large effect that these methodological choices have on polygenic score distributions, see Supplementary Fig. 1 (polygenic scores for height18, weighted with GIANT) and Supplementary Fig. 2 (polygenic scores for PTSD10, from a multi-ancestry GWAS from the Psychiatric Genomics Consortium, PGC).
