Figure 3.

Comparison of imputation quality across experiments for each superpopulation. Variants were binned according to their minor allele frequency in the 1KGP3 and imputation r² averaged across variants in each bin. For all experiments, the 1KGP3 genotypes were treated as “truth” and imputation r² was computed by taking the squared correlation coefficient between the vector of imputed alternate allelic dosages and the truth genotypes. Same results on a log scale are shown in Supplemental Figure 13. Note that imputation performance at low MAF for a given sequencing experiment was often higher in the AFR cohort compared to the EUR cohort.