Accurate haplotype imputation with individualized ancestry-adjusted reference panels

Accurate data imputation requires ethnicity-matched reference panels. However, recent admixtures have created mosaic human genomes, different chromosomal segments have different ethnic backgrounds, so it is impossible for a single-ethnicity reference panel to be the matched for data imputation. In this study, we explored a novel strategy for imputation. We created individualized mosaic reference panel for each person according to his/her ethnic backgrounds at each genomic locus. We examined on datasets with 70% missing values on haplotypes and 50% missing values on genotypes. Results showed that the imputation with mosaic references steadily yielded high imputation accuracy and outperforms the other strategies. With the mosaic reference panels, the imputation accuracy was 98.8 ± 0.1% (CEU), 98.7 ± 0.1% (YRI), 98.5 ± 0.1% (CHB), 98.6 ± 0.1% (ASW), 97.3 ± 0.1% (MKK) and 98.2 ± 0.1% (MXL). Mosaic reference panel will be one option for future missing value imputation in big data era.