We next characterized the source of power gains from the 1000G SNP panel compared with the HapMap SNP panel, focusing on transcripts not associated with SNPs in the HapMap imputation but mapped locally within 1 Mb with at least one 1000G imputed SNP. We paired the peak SNP in the 1000G panel with the peak SNP in the HapMap panel for each transcript and calculated both the linkage disequilibrium (LD) (evaluated using the 1000G haplotypes) between them and the difference in association significance [–log10(P)]. We found most missed hits were due to low LD between HapMap SNPs and the peak SNP identified after 1000G imputation (Supplemental Fig. 4).
