It is useful to run a separate imputation method as an external benchmark. We chose to compare against Beagle (Browning and Browning 2009) because past results showed that it could be competitive on a dataset of this scale (Browning and Browning 2009; Howie et al. 2009; Jostins et al. 2011). Beagle has already been compared with IMPUTE2 in a large, well-matched reference panel of European ancestry (Howie et al. 2009), so to simplify the presentation we applied it only to the cosmopolitan HapMap 3 reference panels. We used Beagle version 3.0.2 with default settings for all experiments presented here. To facilitate parallel computation, we ran Beagle on the same 5-Mb chromosome chunks (with buffers) that were used by IMPUTE2.
