For our benchmarks of phasing without a reference, we created subsets of UK Biobank samples containing N = 5,000, 15,000, 50,000, or 150,000 samples, each of which contained all 72 trio children and none of the 144 trio parents. We then applied each phasing method to phase chromosomes 1, 5, 10, 15, and 20 of each subset of samples, and we computed mean switch error rates and s.e.m. over the 70 European-ancestry trio children as above. We applied an analogous procedure to the GERA sub-cohorts with the same modifications as above.
