MultiCook: A Tool That Improves Accuracy of HLA Imputation by Combining Probabilities From Multiple Reference Panels and Methods

Abstract

HLA molecules are produced by genes within the Major Histocompatibility Complex (MHC). Although the identification of HLA genotype is costly, fortunately, recent computational methods have made it possible to impute HLA genotypes using inexpensive single nucleotide polymorphism (SNP) markers. These imputation methods perform well if ethnicity-matched reference panels are provided. However, the availability of large-sized panels specific to each ethnicity remains limited. As a result, to achieve better imputation for each population, we need to utilise available multiple reference panels together. In this study, we introduce MultiCook, which enables users to simultaneously utilise existing multiple HLA imputation methods with multiple reference panels. MultiCook is versatile in that panels typed with different SNP genotyping platforms can seamlessly be merged, outputs from multiple imputation methods can be combined and the output from the Michigan imputation server with a multiethnic reference panel can also be incorporated. We compared MultiCook to the existing single-reference-panel approaches and the Michigan HLA imputation server. In evaluation with a cohort of 413 Koreans, MultiCook reduced the imputation error rate by about one third, from 4.70% to 3.31%, by combining 1KG EAS (N = 504) and HAN Chinese (N = 9773) reference panels compared to the single-panel approach. Moreover, MultiCook achieved better accuracy for imputing low-frequency alleles in evaluation benchmarks.