Using Haplotype Based Models for Genomic Predictions in Crossbred Animals

We seek to make a haplotype based genomic prediction that can accurately predict tenderness and other meat quality traits in multiple breeds and crossbred animals. We analyzed 651 Angus, 695 Charolais, 1,095 Hereford, 283 Limousin, 301 Maine-Anjou, and 516 Simmental samples with phenotypes and genotypes from the Carcass Merit Project (CMP). We used 3,993 Angus, 101 Charolais, 1,225 Hereford, 2,366 Limousin, 11 Maine-Anjou, and 1,913 Simmental purebred animals along with the CMP animals to phase the genotype data and impute the missing genotypes using BEAGLE v3. Using GEMMA, we fit traits in a Bayesian Sparse Linear Mixed Models (BSLMM). The SNPs with the largest effects are found by calculating allele substitution effect from the SNP based BSLMM analysis. Using the four flanking SNPs of each identified core SNP we construct five SNP blocks that are used to form haplotypes. The five SNP blocks are not allowed to overlap. Using GEMMA, we fit this new haplotype matrix as the random effects in a BSLMM analysis. Using the output from this analysis we predict breeding values using GEMMA. We use a three-fold cross-validation to validate these predictions. We find the correlation and accuracy of each breed and across breeds using the haplotype based models. We are currently evaluating the model to find the optimal way to use haplotypes in genomic predictions.