Predictive ability of genomic selection models in a multi-population perennial ryegrass training set using genotyping-by-sequencing

Perennial ryegrass (Lolium perenne L.) is a key source of nutrition for ruminant livestock in temperate environments worldwide. Higher seasonal and annual yield of herbage dry matter (DMY), a driver of livestock productivity, is a principal breeding objective but the historical realised rate of genetic gain for DMY is modest. Genomic selection was investigated as a tool to enhance the rate of genetic gain for DMY, using genotyping-by-sequencing (GBS) and a multi-population (MP) training set that combined DMY row plot data from five breeding populations phenotyped in five environments over two years. GBS using the ApeKI enzyme yielded more than 1M single nucleotide polymorphism (SNP) markers from the training set of 566 individuals. Cross-validation prediction accuracy (PA) for DMY with the MP-based models ranged from 0.066 to 0.429 and was predominantly higher than within-population models constructed with smaller training sets. Best Linear Unbiased Predictor (BLUP)-based genomic prediction methods, including GBLUP with a either a standard or a recently-developed (KGD) relatedness estimation, were superior or equal to other computational approaches, including Ridge Regression and Random Forest. PA of the genomic prediction models was likely principally an outcome of SNPs modelling genetic relationships between training and validation sets, which has implications for the persistence of PA over generations of selection. However, modelling based on data generated in this study indicated that significant enhancement of genetic gain may be achieved inside a single cycle of selection, if genomic prediction is used for within-family selection at relatively high selection intensity.