Figure 3A shows the results for all imputed SNPs, whereas Figure 3B is restricted to SNPs with MAF < 10%. We omitted SNPs with MAF < 1% from both plots as there were too few of these to provide reliable measurements. Like previous authors (Jallow et al. 2009), we found that a Gambia-specific reference set outperformed a HapMap set that did not include Gambians: the GMB panel (blue curves) produced higher overall accuracy than did the HM3 panel (red curves), despite the fact that there were ∼800 chromosomes of African ancestry in the HM3 set and only 200 chromosomes in the GMB set. This suggests that historical divergence between the populations in the HapMap 3 and Gambian panels makes HapMap 3 less accurate as an imputation resource.
