and 954 controls in AA. Since Linkage Disequilibrium (LD) is an important concern for selecting representative SNPs for each gene, we used Plink http://pngu.mgh.harvard.edu/~purcell/plink/ to prune SNPs that are in strong LD (Plink uses 0.5 as the default pairwise R2 threshold, Supplementary Table 3 in Additional file 1). The final data used in our study consisted of 245,216 SNPs in EA and 482,914 SNPs in AA. The SNPs were assigned to genes on the basis of being located within the gene or up to 20 kb from the gene. Most genes are associated with more than one SNP; we applied the adaptive rank truncated product of SNP association algorithm described in Method to selected representative SNPs for each gene. For pathways, we used 215 experimentally validated pathways from the KEGG database[29] (Release 55, accessed 12 September).
