Genomic prediction using haplotypes in Brown Swiss

In order to improve accuracy of genomic selection different approaches have been suggested. One possibility is to use haplotypes instead of SNPs. It is thought that by the usage of haplotypes the number of effects to estimate should be decreased and the accuracy should be increased because the haplotype should catch the causal variants better than from LD with SNPs. Different definitions of the length of haplotypes are possible. The haplotypes can either be determined by the number of SNPs in a haplotype, by the length in base pairs or by linkage disequilibrium (LD) measures. For this study we used four different definitions of haplotype lengths either based on physical length in bp or on LD measures. We used haplotypes with a length of 250kb or 1Mb, we defined the LD based groups in PLINK and either included or excluded SNPs that were not included in any LD block. We estimated genomic breeding values with each of these haplotype definitions and compared prediction accuracy to that achieved with 50K SNPs for four traits in Brown Swiss. The traits were protein yield, non-return rate 56 in heifers, somatic cell score and stature. Estimation of genomic breeding values was carried out applying a BayesC model. We found trait-specific differences in the ranking of the scenarios. However, differences in accuracies between scenarios within trait were relatively low and using haplotypes only marginally increased the accuracy of genomic breeding values. The number of variables to be fitted increased relative to the SNP model especially for scenarios where the haplotypes were defined by physical length.