Development of Genomic Prediction in Sorghum

Abstract

Genomic selection can increase the rate of genetic gain in plant breeding programs by shortening the breeding cycle. Gain can also be increased through higher selection intensities, as the size of the population available for selection can be increased by predicting performance of non-phenotyped, but genotyped, lines. This paper demonstrates the application of genomic prediction in a sorghum [Sorghum bicolor (L.) Moench] breeding program and compares different genomic prediction models incorporating relationship information derived from molecular markers and pedigree information. In cross-validation, the models using marker-based relationships had higher selection accuracy than the selection accuracy for models that used pedigree-based relationships. It was demonstrated that genotypes that have not been included in the trials could be predicted quite accurately using marker information alone. The accuracy of prediction declined as the genomic relationship of the predicted individual to the training population declined. We also demonstrate that the accuracy of genomic breeding values from the prediction error variance derived from the mixed model equations is a useful indicator of the accuracy of prediction. This will be useful to plant breeders, as the accuracy of the genomic predictions can be assessed with confidence before phenotypes are available. Four distinct environments were studied and shown to perform very differently with respect to the accuracy of predictions and the composition of estimated breeding values. This paper shows that there is considerable potential for sorghum breeding programs to benefit from the implementation of genomic selection.