Performance of genetic imputation across commercial crop species

Abstract

We show that accurate imputation can be carried out in three commercial plant species (maize, sugar beet and wheat) and that accurate imputation does not require a pedigree, although pedigree information can improve accuracy and speed. Our approach uses a hidden Markov model to build a haplotype library from individuals genotyped at high-density and then uses this library to impute low-density genotyped individuals to high-density. To build the library, we use founders when the pedigree is known, or a sample of progeny when the pedigree is unknown. Without a pedigree, and with 50 individuals genotyped at high-density and 100 low-density markers per chromosome, the median accuracies were 0.97 (maize), 0.96 (sugar beet), and 0.94 (wheat). We obtained similar accuracies with a pedigree. For biparental crosses with 100 markers per chromosome, median accuracies were 0.96 (maize), 0.96 (sugar beet) and 0.94 (wheat). For the imputation scenarios without a pedigree, we compared accuracies with those obtained by running Beagle 5.1. In all but one scenario, our method outperformed Beagle. We believe that plant breeders can effectively apply imputation in many crop species.