We benchmarked Eagle113, Eagle2 (v2.1), SHAPEIT v2 (r790)12, and Beagle v4.1 (22Feb16.8ef)25 using the Oxford genetic map (supplied with SHAPEIT and Eagle). When running Eagle2 in reference-based phasing mode, we turned off imputation of missing genotypes using the –noImpMissing flag; otherwise, we ran all methods using their default parameter settings unless explicitly testing non-default settings. Specifically, the non-default parameter settings we tested were the –no–mcmc and –states (K) option of SHAPEIT2 and the –Kpbwt option of Eagle2.
