The dimensions of genomics datasets used in breeding programs are rapidly increasing. CRV currently has a reference population of ~30,000 Holstein Friesian dairy bulls, partly bulls previously tested in the Netherlands, and partly foreign bulls made available through the EuroGenomics consortium. Furthermore, CRV has smaller reference populations consisting of 1,300 – 2,500 progeny tested bulls, for Jerseys and Friesians held under specific circumstances (“grazing” in New Zealand). Further expansion of those reference populations by genotyping large numbers of cows that are phenotyped for a large number of traits is currently underway. Several issues that are important to make the most out of this great resource are addressed in the project GenomXL.
One important restriction for routine genetic evaluations, is that the runtime should be limited, such that breeding values can be estimated rapidly. This requirement becomes a challenge when tens of thousands of animals are genotyped, and also when in the near future imputed sequence data with many millions of SNPs may be available for those animals. In GenomXL, two important steps forward have been made in this regard. Firstly, a new algorithm is developed to estimate SNP effects. This algorithm is called “right-hand-side” updating, and was shown to be able to reduce CPU time by 74.5 to 93.0% and memory requirements by 13.1 to 66.4% (Calus, 2014). Secondly, it has been shown that genomic evaluations can use previously estimated SNP variances, such that genomic evaluations can be run using fast converging BLUP models, while capturing properties of Bayesian variable selection models (Calus et al., 2014).
Multi-breed genomic prediction
Using the large Holstein Friesian reference population, reliable breeding values can be predicted for Holstein Friesian animals. Other breeds, like Jerseys or local breeds, lack this large reference population, resulting in less reliable breeding values. In GenomXL, the potential to use genomic information across breeds is investigated. As a first step, different factors determining the accuracy of genomic prediction were investigated, showing a large effect of family relationships and a much smaller effect of linkage disequilibrium between markers and genes (Wientjes et al., 2013). In the next step, the potential accuracy of across and multi breed genomic prediction was investigated. Results showed that, due to the lack of close family relationships across breeds, accuracy of across breed genomic prediction is low compared to the accuracy achieved for single breed genomic prediction, even when gene effects are the same (Wientjes et al., 2014a). Differences in gene effects reduce the potential accuracy, proportional to the correlation between gene effects across populations (Wientjes et al., 2014a). Moreover, accuracy is depending on the genetic architecture of the trait; when allele frequency of genes underlying a trait decrease, accuracies of both single and multi-breed genomic prediction decrease as well (Wientjes et al., 2014b).