Nonetheless, when using the February 2012 release of 1000 Genomes, imputation quality for low frequency SNPs was highest when using the most closely related reference panel (average r2hat = 0.51 when using YRI+CEU+ASW, 0.45 when using AFR+EUR, and 0.37 when using ALL, for SNPs with MAF≤2%). For SNPs in the remainder of the MAF spectrum, the highest imputation quality was observed when using the most diverse reference panel (average r2hat = 0.83 when using YRI+CEU+ASW and 0.86 when using either EUR+AFR or ALL, for SNPs with MAF>2%). This pattern resulted from poor imputation quality for low frequency SNPs in the ALL panel that are monomorphic in the more closely related panels. The pattern of reduced overall imputation quality in more diverse panels was similarly observed for the MaCH programs (Figure 3), due to poor imputation quality of low frequency SNPs (Figure S6 for MaCH and Figure S7 for MaCH-Admix). When using BEAGLE, imputation quality of low frequency SNPs was not negatively affected by the inclusion of more distantly related populations in the AFR+EUR and ALL panels (Figure 3 and Figure S8).
