We considered restricting to SNPs imputed with increasingly higher quality, as quantified by the imputation R2. Results for the soft call genotypes are shown in Table 1. As the R2 threshold was increased, our summary measures improved; however, this happened at the expense of losing SNPs for analysis, which reflects some loss of power. It should also be noted that even at the most stringent threshold listed, R2 > 0.99, when we’ve excluded nearly 70% of the SNPs, there remain 57 SNPs with p < 5 × 10−8. Figure 3 shows the discriminatory ability of this method as we vary the R2 threshold.
