Using the same Illumina 1M versus 550v3 comparisons, we evaluated whether the imputation-induced bias based on the union of genotyped SNPs differed according to high versus low LD patterns. We employed the LD pruning procedure in PLINK (Purcell et al. 2007) with a 1,500 SNP sliding window to select SNPs under high LD with other SNPs (squared correlation coefficient between SNPs [r2] >0.8) and conversely SNPs in linkage equilibrium (r2 < 0.2). SNPs were selected according to the LD patterns in the SAGE subjects, separately by ethnic group, and their SNP association results were taken from the comparison to PanScan for European Africans (Fig. 1b) or iControl for African Americans (Fig. 1d). In the European Americans, 0.64 % of the 2,984 imputed SNPs in low LD and 0.18 % of 23,072 imputed SNPs in high LD were false positives. In African Americans, 0.30 % of the 6,590 imputed SNPs in low LD and 0.18 % of 19,134 imputed SNPs in high LD were false positives.
