Training population selection for (breeding value) prediction

Abstract: Training population selection for genomic selection has captured a great deal of interest in animal and plant breeding. In this article, we derive a computationally efficient statistic to measure the reliability of estimates of genetic breeding values for a fixed set of genotypes based on a given training set of genotypes and phenotypes. We adopt a genetic algorithm scheme to find a training set of certain size from a larger set of candidate genotypes that optimizes this reliability measure. Our results show that, compared to a random sample of the same size, phenotyping individuals selected by our method results in models with better accuracies. We implement the proposed training selection methodology on four data sets, namely, the arabidopsis, wheat, rice and the maize data sets. Our results indicate that dynamic model building process which uses genotypes of the individuals in the test sample into account while selecting the training individuals improves the performance of GS models.