Weighted kernels improve multi-environment genomic prediction.

Area of Science:

• Plant breeding
• Quantitative genetics
• Genomics

Background:

• Genomic selection (GS) is vital for variety improvement, but single-environment models struggle with genotype by environment (G×E) interactions.
• Current GS models often estimate genomic similarity rather than population genetic characteristics, limiting predictive power.

Purpose of the Study:

• To develop an advanced GS framework integrating genetic associations and population characteristics for improved cross-environment prediction.
• To enhance the accuracy of predicting genotype performance across diverse environments.

Main Methods:

• Developed a novel GS framework utilizing multi-environment weighted kernels.
• Incorporated signals from genetic associations of phenotypic variation and population genetic characteristics.
• Employed a Bayesian implementation for flexibility and generalizability.

Main Results:

• Achieved up to a 31% gain in prediction predictability for winter wheat doubled haploid (DH) populations.
• Demonstrated a 4-33% improvement in prediction accuracy for half-sib families compared to Gaussian kernels.
• Validated the framework's capacity to predict performance across environments for genetically heterogeneous populations.

Conclusions:

• The proposed GS framework significantly enhances prediction accuracy across environments by leveraging genetic associations.
• This approach offers a more robust and generalizable tool for plant breeding programs, even with non-uniform genetic materials.
• The multi-environment weighted kernels provide a superior alternative to Gaussian kernels for capturing complex genetic interactions.