For each dataset, unrelated subjects were subset into the three ancestry groups (EUA, AFA, AMA; Supplementary Tables 3, 5, 6) for analysis. SNPs were excluded that had a MAF <5%, HWE P > 1 × 10−3, call rate <98%, were ambiguous (A/T, G/C), or due to being located in the MHC region (chr. 6, 25–35 MB) or chromosome 8 inversion (chr. 8, 7–13 MB). SNPs were pairwise LD-pruned (r2 > 0.2) and a random set of 100 K markers was used for each subset to calculate PC’s based on the smartPCA algorithm in EIGENSTRAT70.
