Evaluation of Genetic Effects on Competition to Improve Genomic Prediction in Cassava Field Experiments

The rate of genetic advancement in plant breeding experiments can be increased by accounting for interference or competition caused by neighboring individuals. Genomic selection (GS) models can be designed that simultaneously predict performance of an individual and the competition of its neighbors. Such models were applied to cassava (Manihot esculenta) breeding experiments conducted in collaboration with International Institute of Tropical Agriculture (IITA). GS models can be used to predict genetic value for selecting parents by taking advantage of the relationship among genotypes through genomic relationship matrix. Three types of incidence matrices were used to relate neighboring genotype to the performance of a focal plot in GS models to account for competitive effects on cassava root yields while simultaneously estimating the direct genetic effects on those yields. The competition incidence matrices considered either only the nearest neighbors or neighbors further away where competition effects were assumed to decay in a slow or fast rate. Predictive ability of candidate models were tested through a 10-fold cross-validation method which was repeated five times. The best model was chosen as the one with the lowest prediction root mean squared error (pRMSE) compared to a base model having only the direct genetic effect component. We found that in most instances, competition stopped at the nearest neighbor. In instances where it went beyond the nearest neighbor, the competition effect decayed slowly mostly. A decrease in pRMSE up to 2.8% was found when competition was accounted for in GS models.