In the wake of these results, we suspect that the accuracy improvement of IMPUTE v2 over IMPUTE v1 is not practically meaningful for imputation based on the HapMap Phase II data. However, given that IMPUTE v1's computational requirements scale quadratically with the number of chromosomes in the reference panel while IMPUTE v2's requirements grow linearly, the newer version may become more computationally favorable as baseline reference panels grow in the future. For example, expanding the HapMap reference panel in this study to 800 chromosomes (which is roughly the size anticipated for each panel in the 1,000 Genomes Project) would lead to similar running times for both versions of IMPUTE, but version 2 would need only 2% of the computer memory required by version 1. At the same time, IMPUTE v2 would probably achieve higher accuracy, and its computational advantages over IMPUTE v1 would continue to grow with larger reference panels.
