We first benchmarked Eagle2 against previous reference-based phasing methods using reference panels generated by phasing subsets of genotyped cohorts. These benchmarks allowed us to explore a greater range of reference panel sizes and genetic ancestries than currently available in sequenced reference panels (as the N=32,470 samples currently in the HRC are predominantly European), understanding that genotyped reference panels containing a limited set of markers are not broadly useful for reference-based phasing. We performed benchmarks using a total of five data sets: the UK Biobank cohort22 and the four GERA sub-cohorts, which were genotyped on four distinct European, African, East Asian, and Latino genotyping arrays23,24. All five data sets were typed on arrays containing 650K–850K autosomal markers with typical heterozygosity and missingness rates, and each data set contained a small subset of mother-father-child trios (Online Methods and Supplementary Table 1).
