Training Population Optimization for Genomic Selection in Miscanthus

Miscanthus is a C4 perennial grass with great potential for lignocellulosic ethanol biofuel production. However, there is a significant drawback for the further development of this biofuel crop due to lack of sufficient winterhardiness in northern latitudes. Abundant genetic diversity exits for different traits in this crop species that can be introgressed to improve current cultivars. In this study we explored the use of diversity panels for training genomic selection (GS) models to predict phenotypically optimal lines. Also, we examined the challenges associated such methodology by (1) evaluating the impact of population structure in Msi and Msa diversity panels and (2) quantify the advantages and disadvantages of using both Msi and Msa panels as a training set for fitting GS models. Discriminant analysis of principal components, principal component analysis, and coefficient of determination methodologies were used to assess the impact of population structure on the prediction accuracy of the GS models. Our results demonstrated that population structure did have an effect on the performance of GS models, where the performance varied among the methods employed to account for population structure. GS prediction accuracies varied across training sets from low to moderate and were trait dependent. The implications of the results of this study, in particular the considerations one needs to take when fine tuning training sets and GS models for optimal predictions of breeding values, will be discussed.