Evaluation of strategies to optimize training populations for genomic prediction in oat (Avena sativa)

Abstract

Genomic selection is a promising breeding methodology that could increase selection accuracy and intensity and reduce generation interval. As the cost of genotyping decreases, it will be important to optimize training populations for costly phenotypic experiments for many complex traits. The aim of this research was to evaluate different optimization strategies, by using historical data from the Norwegian oat breeding programme at Graminor. In this paper, we focus on the optimization criteria: genetic diversity, phenotypic variance and genetic similarity between the training and testing populations. The four training population strategies—prediction core, diversity core, phenotypic selection and random selection—were applied to an oat candidate population of 1124 lines. An independent testing population was used to calculate the mean prediction abilities for the traits days to heading and plant height. Moreover, the strategies were tested in three independent wheat populations. The results showed that prediction core was the most promising strategy to select training populations with high genetic similarity to the testing set, high genetic diversity, and high phenotypic variance, which resulted in higher prediction ability across population sizes and traits.