New Developments in Plant Genomic Prediction Models

We examine results from three areas of research that aim to increase genomic prediction accuracy in plant breeding: (1) inverse penalized regression models, (2) prediction models that incorporate genotype × environment interaction, and (3) threshold models for ordinal traits commonly used in plant breeding for assessing various plant diseases. Recently, models were developed within the framework of inverse regression theory. Inverse solutions make it possible to visualize that noise is inversely proportional to singular values and graphics depict the shrinkage behavior according to the prior variance and weighting factor. Inverse Bayesian regression models deal well with the ill-conditioning and random noise problems arising in genomic prediction when p>>n. Results show that inverse Bayesian regression models increase prediction accuracy compared with standard Bayesian regression models. Prediction models incorporating genotype × environment interaction (GE) have been developed as extensions of the GBLUP. These models allow incorporating highly dimensional marker and environmental covariable data and increase prediction accuracy when incorporating GE. Another type of GBLUP model uses marker x environment interaction for modeling GE. This model provides produced substantial gains in prediction accuracy, provided that the environments analyzed are positively correlated. Threshold genomic predictions model for ordered categorical plant disease data incorporating GE and addtive x additive interactions indicated increased in prediction ability as compared with GBLUP threshold model without GE and addtive x additive interactions.