Kernel-based whole-genome prediction of complex traits: a review

Area of Science:

• Quantitative genetics
• Genomics
• Statistical modeling

Background:

• Whole genome-enabled prediction leverages high-dimensional genomic data and molecular markers.
• Traditional models are extended to address challenges posed by complex traits and interactions.
• Kernel methods are increasingly favored for genome-enabled prediction due to their ability to model complex relationships.

Purpose of the Study:

• To review whole-genome regression using kernel methods for quantitative traits in agriculture.
• To discuss kernel-based approaches for capturing total genetic variation and enhancing predictive performance.
• To explore connections between prediction models in animal breeding, statistics, and machine learning.

Main Methods:

• Review of kernel-based statistical approaches for genome-enabled prediction.
• Focus on methods capturing non-additive genetic effects (parametrically and non-parametrically).
• Analysis of whole-genome regression models incorporating genome annotation information.

Main Results:

• Kernel methods show potential for capturing non-additive genetic effects.
• Encouraging results have been observed with non-parametric kernel approaches.
• Integrating genome annotation can potentially improve predictive performance.

Conclusions:

• Kernel methods offer a promising avenue for improving prediction of genetic values, especially for complex traits.
• Capturing non-additive genetic variation remains a significant challenge in quantitative genetics.
• Further research is needed to fully exploit kernel methods for accurate genetic predictions.