Genotype Imputation with Population-Specific Reference Panels

• Activity Number: 526
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Hye Sung Min*+ and Matthew Zawistowski and Sebastian Zöllner
• Companies: and University of Michigan and University of Michigan
• Keywords: coalescent ; imputation ; sequencing

Abstract

Genotype imputation is a common method for estimating genotypes at unobserved markers. Typical algorithms of a Hidden Markov Model compare haplotypes between a target set and more densely typed reference set and copy to each target haplotypes from the best matching template. Since genetic similarity of reference and target samples affects imputation accuracy, selecting those two haplotypes from the same population is desirable. In practice, reference and target samples are derived from similar, but distinct population. However, the recent reduction in sequencing costs allows creation of population-specific reference panels. Here, we evaluate the benefit of such reference samples using a coalescent-based model. We consider reference and target samples simulated with varying levels of divergence. After masking most genotypes in the target sample, we impute these genotypes using MaCH. In results, population structure affects both the quantity of sites that can be imputed in the study sample and imputation accuracy. Common variants can be properly imputed from divergent reference panels; however, rare variant imputation significantly benefits from population-specific reference panels.