Genomic Prediction of Genotype × Environment Interaction Kernel Regression Models.

Area of Science:

• Agricultural Science
• Genetics
• Biotechnology

Background:

• Genomic selection (GS) is crucial for crop improvement.
• Genotype × Environment interaction (G × E) significantly impacts breeding success.
• Modeling G × E, particularly marker × environment (M × E) interactions, is essential for accurate predictions.

Purpose of the Study:

• To propose and evaluate two novel nonlinear Gaussian kernel methods for modeling G × E in GS.
• To compare the performance of these methods (RKHS KA, RKHS EB) against traditional linear kernel approaches (GBLUP).
• To assess prediction accuracy in single-environment and across-environment analyses for wheat and maize.

Main Methods:

• Implementation of reproducing kernel Hilbert space with kernel averaging (RKHS KA).
• Application of Gaussian kernel with empirical Bayesian bandwidth estimation (RKHS EB).
• Comparison with single-environment GBLUP and GBLUP-G × E models using wheat and maize datasets.

Main Results:

• Both RKHS EB and RKHS KA demonstrated higher prediction accuracy than GBLUP in single-environment analyses.
• RKHS KA-G × E and RKHS EB-G × E showed substantial superiority (60-68%) over single-environment models for wheat under positive correlations.
• Gaussian kernel models achieved higher accuracies (up to 17% for wheat, 5-6% for maize) compared to GBLUP-G × E.

Conclusions:

• Nonlinear Gaussian kernels (RKHS KA, RKHS EB) offer superior prediction accuracy in GS compared to linear kernels.
• The flexibility of Gaussian kernels effectively captures complex marker main effects and marker-specific interactions.
• These advanced models hold significant promise for enhancing crop breeding programs by improving genomic prediction accuracy.