The subsequent subject-level QC procedures are outlined in Figure S1. First, we verified that all iControlDB African Americans had call rates>95%. Identity-by-state (IBS) estimates were then calculated to identify possibly duplicated subjects. For pairs of subjects having IBS>99%, we retained the subjects with the highest call rate from each pair. Identity-by-descent (IBD) estimates are often generated from GWAS data to remove subjects with cryptic relatedness, whose inclusion would violate independence assumptions in subsequent statistical analyses. However, population stratification may inflate these estimates, so we used the KING program, which was designed to circumvent the inflation of IBD estimates due to stratification [20]. We used a kinship coefficient threshold of 0.0441 to identify clusters of third-degree or closer relatives [20], and we retained only one subject having the highest call rate from each relative cluster. Following the aforementioned QC criteria, 774 (93.3%) subjects remained. None of these subjects had excessive homozygosity or discordance between reported gender and estimated gender based on chromosome X SNP data.
