We recently compared several methods all designed to phase nominally unrelated samples (SHAPEIT2 [2], SHAPEIT1 [5], Beagle [4], HAPI-UR [6], Impute2 [24], MaCH [25], fastPHASE [26]) and found that SHAPEIT2 was the most accurate method in this setting. In this paper we examine the performance of these methods at increasing levels of relatedness between individuals. To do this we used cohorts from six different isolated populations (and two additional cohorts from non-isolated populations). Each of these cohorts contain some extended pedigrees allowing us to assess performance on both nominally unrelated individuals and on explicitly related samples.
