One observation from Figure 1 is that the full set of reference haplotypes generated some of the highest accuracy levels in this experiment – the black curves almost always lie above the other curves. As we show in Figure 2 and File S3, this holds true across HapMap 3 target panels, genotyping arrays, and SNP frequency classes. Another salient feature of Figure 1 is that the curves plateau quickly with increasing values of khap (moving from left to right within each plot). This shows that our surrogate family approximation can decrease computing time without losing accuracy. For example, the runs that selected 500 haplotypes from the full HapMap 3 panel (khap = 500; black curves) achieved similar accuracy to the runs that didn't use this approximation (khap = 2020), but the first set of runs was about four times faster.
