When calculating SE rate across the whole of chromosome 10 (not just in IBD regions), SHAPEIT2 also has the lowest error rate, ranging from 0.28% (GPC) to 2.65% (Split cohort) as opposed to 0.49% and 5.57% for Beagle and 0.50% and 11.10% for HAPI-UR. Switch error rates for SLRP cannot be evaluated across the whole of chromosome 10 due to the method only producing partially phased haplotypes. We observe that all methods perform relatively better within IBD regions than across the whole chromosome. However, the difference for SHAPEIT2 is much larger than for Beagle and HAPI-UR. These results suggest that whilst none of these methods explicitly model IBD sharing, its presence tends to be exploited implicitly, and particularly so in SHAPEIT2. Figure S6 (right) plots the SHAPEIT2 SE rate within IBD regions (detected by SLRP) against the rate outside these regions. Switch error is clearly close to zero when IBD sharing is present and has a rate more comparable to non-isolated populations when no IBD sharing is present.
