The x-axis shows the value of the khap parameter, and the y-axis shows the imputation accuracy, which is measured as the mean SNP-wise R2 between true and imputed allele dosages (posterior mean genotypes) for each cross-validation experiment. The computational cost of imputation is roughly proportional to khap. Applying different khap settings to a single reference panel generates a curve, and each curve represents a different reference panel. (For further details of the R2 distributions underlying the mean values at khap = 500, see File S2.) Each individual was imputed initially from haplotypes in the same panel, then from reference sets that cumulatively added panels in the order shown in the plot legends, reading from bottom to top. The black curves represent our suggested strategy of using a cosmopolitan reference panel.
