Throughout our imputation experiments, we found that IMPUTE2 can attain both higher accuracy and faster computation than Beagle, which is a leading inference method for large datasets. We believe that the success of IMPUTE2 in this context can be attributed to its computational strategies and its model of DNA sequence variation. We discuss these attributes here in hopes that they will inform future methods development, and we address the role that prephasing can play in speeding up imputation.
