We performed genome-wide association analysis separately in EAs and AAs first, and then performed a meta-analysis of EAs and AAs. Genome-wide false discovery rates (FDRs)[Benjamini and Hochberg 1995] were calculated in EAs, AAs and meta-analysis. The replicable risk regions were identified, in which (1) many markers were associated with phenotype across EAs, AAs and meta-analysis, and (2) the distributions of -log(p) values for associations for all markers were consistent across EAs, AAs and meta-analysis; that is, the number of risk markers, the effect directions, the effect sizes and the significance strengths were congruent across three groups. These distributions were compared for similarity using Pearson correlations. Among these replicable risk regions, those with p<5×10-7 and FDR<0.05 in the meta-analysis were selected as “genome-wide significant and replicable regions” (screening step; see correction for multiple testing section below). And then, all SNPs in those significant and replicable regions were examined to identify all replicable risk SNPs (Table 1) (testing step). Also, around these significant and replicable regions, a region spanning 5Mb to whole chromosome was carefully studied, to know the width of the selected risk region where the putative causal loci may be located.
