We have developed a coherent and convenient imputation framework that addresses these concerns. To simplify the process of choosing reference haplotypes from a diverse collection, our approach uses a cosmopolitan reference panel. Previous work suggested that cosmopolitan panels could increase imputation accuracy at low-frequency variants, and our results extend these findings to a wide range of human populations. It is computationally intensive to perform imputation with large, ancestrally diverse reference panels, so we have also developed an approximation that decreases the cost of adding haplotypes to a reference set without sacrificing accuracy.
