We addressed this question by performing cross-validation experiments in Gambian individuals from MalariaGEN. In concurrence with a more limited analysis by Jallow et al. (2009), we found that a population-specific reference panel yielded higher average accuracy than did a larger HapMap 3 panel that lacked Gambian haplotypes. However, we also found that the HapMap 3 panel produced reasonable imputation accuracy across the allele frequency spectrum, and that non-Gambian haplotypes improved accuracy when added to a Gambia-specific reference panel.
