Using eight cohorts from isolated and non-isolated populations, all containing explicitly related individuals, we have carried out a comprehensive evaluation of approaches for haplotype estimation in the presence of IBD sharing. We compared approaches that are specifically focused on estimation of haplotypes in isolated samples (SLRP) and others (SHAPEIT2, Beagle and HAPI-UR) that were designed predominantly for cohorts of nominally unrelated individuals. Our experiments show that the SHAPEIT2 method provides high quality haplotypes that are more accurate than those estimated by SLRP, whereas Beagle and HAPI-UR produce results that are worse than SLRP. We find that the SE rates of SHAPEIT2 are a fraction of a percent in all cohorts, whereas the approaches BEAGLE and HAPI-UR produce SEs that are an order of magnitude larger.
