By using both the reference panel and the study sample to inform phasing updates in Step 1, IMPUTE v2 uses more of the information in the data than most comparable methods [6],[11], which typically account for phase uncertainty using only the reference panel. At the same time, each iteration is relatively fast because untyped SNPs are imputed in a haploid framework rather than the more computationally intensive diploid framework that is used by other HMM methods. For example, one iteration of IMPUTE v2 will typically finish faster and use less computer memory than a run of IMPUTE v1 on the same dataset, although IMPUTE v2 tends to be slower than IMPUTE v1 on the whole since the new method requires multiple iterations. We explore the computational burden of the method in detail in the Results section.
