Genotype Imputation Accuracy Between Low- and High-Density Equine SNP Arrays

The development of 54K and 65K SNP genotyping arrays has lead to successful mapping of loci contributing to many equine health and performance traits using genome-wide association and/or identification of genomic signatures of selection. However, the SNP density of the first-generation arrays has proven inadequate for mapping complex traits and for mapping in breeds with high diversity and rapid decay of linkage disequilibrium. Genotype imputation utilizes dense reference haplotype information to computationally infer unknown genotypes in a more sparsely genotyped test population. We have created a haplotype resource using data from the recently released Affymetrix Equine MNEc2M SNP chip, with which 54K and 65K SNP data can be imputed. Our objective was to assess imputation performance in five horse breeds and develop recommended parameters for post-imputation variant filtration. Using a reference population of 516 horses from 32 breeds genotyped for ~1.8 million SNPs, test populations of Belgian (n = 72), Morgan (n = 285), Quarter Horse (n = 228), Standardbred (n = 475), and Thoroughbred (n = 491) horses were imputed using BEAGLE4 and IMPUTE2. In each test population, 15-30 horses were also present in the reference population, allowing an assessment of the accuracy of genotypes called by the programs. Concordance rates for IMPUTE2 and BEAGLE4 ranged from 99.1-99.8%. These data will next be used to determine post-imputation SNP filtration criteria to maximize confidence in genotypes called in test populations for which there are no individuals represented in the reference panel.