We imputed our simulated GWAS dataset from two different reference panels: a “Cosmopolitan” panel containing the full set of 4800 haplotypes and a “European” panel containing 1000 haplotypes sampled from the 1600 CEU-like reference chromosomes. The first panel models the worldwide 1000 Genomes set, whereas the second panel models a single cluster of related populations (roughly speaking, the 1000 Genomes samples will be divided among five such clusters containing 500 individuals each). By imputing from both reference panels, we can see how reference panel size and diversity affect the computational loads of different imputation methods.
