Selective Genotyping and Cross-Validation Strategies in Genomic Selection

Genome-wide marker information can improve the reliability of breeding value predictions for young selection candidates within the so-called genomic selection. However, the cost of commercial genotyping may limit its use to elite animals, and how such selective genotyping approach affects predictive ability of genomic selection models is still an open question. In addition, it is known that the level of genetic relatedness between animals in the reference population, in which the genomic selection model is trained, and animals for which genetic merit will be predicted using such a model has an important impact on accuracy of genome-enabled predictions. In this presentation, we will discuss these two issues and illustrate their implications using some simulated and real data examples. Some competing selective genotyping strategies will considered, as well as alternative setting for performing cross-validation within a genomic selection context. It will be shown that considerable increments in prediction accuracy can be obtained with carefully chosen animals to be genotyped, and that alternative cross-validation layouts can better indicate the expected prediction accuracies of different applications of the genomic selection model estimated.