Genomic-Enabled Prediction Kernel Models with Random Intercepts for Multi-environment Trials.

Area of Science:

• Quantitative genetics
• Plant breeding
• Genomic prediction

Background:

• Genotype-by-environment (G×E) interactions are crucial for accurate genomic prediction.
• Modeling G×E interactions effectively is challenging, especially with complex environmental correlations.

Purpose of the Study:

• To compare the prediction accuracy of various genomic prediction models with and without G×E interactions.
• To evaluate the impact of incorporating random intercepts for lines on model performance.
• To assess model efficiency and accuracy across maize and wheat datasets with differing environmental correlations.

Main Methods:

• Compared 16 model-method combinations, including main genotypic effect, G×E deviation models (MDs, MDe), and unstructured variance-covariance models (MUC).
• Utilized linear kernel (Genomic Best Linear Unbiased Predictor, GB) and Gaussian kernel (GK) methods.
• Incorporated random intercepts for lines to model genetic residuals.
• Tested models on maize datasets (positive phenotypic correlations) and wheat datasets (complex G×E with negative/zero correlations).

Main Results:

• For maize, MDs and MDe models with random intercepts and GK method showed computational efficiency and high prediction accuracy.
• For wheat, MDs and MDe models with random intercepts and GK method offered significant computational savings over MUC models.
• However, for wheat, these efficient models exhibited slightly lower genomic prediction accuracy compared to MUC models.

Conclusions:

• Models incorporating G×E interactions and random intercepts are computationally efficient and accurate for maize.
• For complex G×E scenarios like wheat, these models provide a balance between computational efficiency and prediction accuracy.
• The choice of model and method should consider the specific crop and the nature of G×E interactions.