Accuracy and Training Population Design for Genomic Selection on Quantitative Traits in Elite North American Oats

Abstract

Genomic selection (GS) is a method to estimate the breeding values of individuals by using markers throughout the genome. We evaluated the accuracies of GS using data from five traits on 446 oat ( L.) lines genotyped with 1005 Diversity Array Technology (DArT) markers and two GS methods (ridge regression–best linear unbiased prediction [RR-BLUP] and BayesCπ) under various training designs. Our objectives were to (i) determine accuracy under increasing marker density and training population size, (ii) assess accuracies when data is divided over time, and (iii) examine accuracy in the presence of population structure. Accuracy increased as the number of markers and training size become larger. Including older lines in the training population increased or maintained accuracy, indicating that older generations retained information useful for predicting validation populations. The presence of population structure affected accuracy: when training and validation subpopulations were closely related accuracy was greater than when they were distantly related, implying that linkage disequilibrium (LD) relationships changed across subpopulations. Across many scenarios involving large training populations, the accuracy of BayesCπ and RR-BLUP did not differ. This empirical study provided evidence regarding the application of GS to hasten the delivery of cultivars through the use of inexpensive and abundant molecular markers available to the public sector.