Bayesian Genomic Prediction with Genotype × Environment Interaction Kernel Models.

Area of Science:

• Plant Breeding
• Genetics
• Statistical Genomics

Background:

• Genotype × environment (G × E) interactions reduce selection accuracy and genetic gains in plant breeding.
• Genomic prediction models incorporating G × E are crucial for improving breeding programs.
• Existing multi-environment G × E models have limitations.

Purpose of the Study:

• To propose two novel multi-environment Bayesian genomic models to better capture G × E interactions.
• To evaluate the prediction ability of these new models compared to single-environment models.

Main Methods:

• Developed two multi-environment Bayesian genomic models using Kronecker products of variance-covariance matrices and genomic kernels (linear/GBLUP and Gaussian/GK).
• The second model includes an additional random effect component (F) to capture unexplained environmental variation.
• Validated models using five CIMMYT datasets (maize and wheat).

Main Results:

• Multi-environment models incorporating G × E consistently outperformed single-environment models in prediction ability.
• The model with the additional random effect (F) showed higher prediction ability than the model without it 85% of the time (GBLUP) and 45% of the time (GK).
• The random effect 'f' provides additional benefits for prediction ability.

Conclusions:

• The proposed multi-environment Bayesian genomic models effectively capture G × E interactions.
• Including the additional random effect 'f' further enhances prediction ability in genomic selection.
• These models offer improved tools for plant breeding to achieve higher genetic gains.