Prediction of Causative Genomic Relationships Using Sequence Data of Five French and Danish Dairy Cattle Breeds

The increasing number of sequenced individuals makes the use of sequences for genomic prediction feasible. With the current SNP chips, the accuracy of genomic selection is limited in populations with low linkage disequilibrium, as is the case in across-breed prediction in dairy cattle. Sequence data is more likely to contain causative mutations and therefore increase the prediction accuracy in such populations. We studied the potential advantage of using real sequence data for prediction of genomic relationships at causative mutations using sequence data of chromosome 1 for 122 Holstein, 27 Jersey, 28 Montbéliarde, 23 Normande and 24 Danish Red bulls. Different scenarios varying the number of causative mutations (10, 50, 100 or 250), minor allele frequency of causative mutations and prediction markers, and sets of prediction markers were used. Prediction markers were either SNP on the 50K and HD chips, or two 1 Kb intervals on both sides of each causative mutation, varying the distance between causative mutations and intervals from 1 base to 1 Mb. Subsequently, the regression coefficient of the genomic relationships at prediction markers on the genomic relationships at causal loci was used to quantify the reduction in prediction R² resulting from the use of prediction markers rather than the causative mutations. The reduction in prediction R² was lowest when a limited number of variants close to the causative mutations was used and increased when more variants were added or the distance to the causative mutations increased. This increase was larger across breed than within breed.