Machine Learning for Genomic Prediction in Lolium perenne

Genomic selection (GS) is an approach to breeding whereby selection decisions are based on using statistics to form genomic estimated breeding values (GEBVs), rather than the traditional approach of progeny assessment or kinship data. Thanks to recent advances in genotyping techniques, and the ease of obtaining ever larger numbers of high-density markers, there is currently increasing interest in the technique, and much progress has been made applying it to animal and plant breeding. However, GEBVs are generally estimated using classical linear models, and their various extensions. Here we present results based on machine learning (ML) techniques that yield themselves well to use in GS and demonstrate their application to predicting quantitative traits in a breeding population of Lolium perenne. We will present the results of this and how they compare to other methods for a range of quantitative traits. We also explore two methods of combining one or more inhomogeneous sub-populations to improve results. We conclude that some ML approaches can successfully compete with classical methods, and can surpass them in some cases.