GWKBR: a novel method integrating machine learning and Bayesian inference framework to improve genomic prediction...

Area of Science:

• Quantitative genetics
• Machine learning
• Bioinformatics

Background:

• Non-additive genetic effects complicate traditional genomic prediction.
• Kernel-based machine learning and Bayesian methods show promise in capturing complex genetic interactions.

Purpose of the Study:

• To develop a novel genomic prediction method, GWKBR, integrating machine learning and Bayesian inference.
• To evaluate GWKBR's performance against existing methods for genomic prediction accuracy.

Main Methods:

• Developed Genome-Wide Association Studies-Weighted Gaussian Kernel Bayesian Regression (GWKBR).
• Integrated weighted Gaussian kernel regression, Bayesian optimization, Bayesian inference, REML, GWAS, and cross-validation.
• Constructed a weighted Gaussian kernel to capture non-additive effects and SNP importance.

Main Results:

• GWKBR demonstrated robust genomic prediction accuracy across simulated, human, plant, and animal datasets.
• The method showed particular advantage in plant datasets influenced by non-additive effects.
• GWKBR achieved the highest average accuracy for 13 out of 23 traits and second-highest for seven.

Conclusions:

• GWKBR offers a reliable and robust approach for genomic prediction in diverse species.
• The method effectively handles diverse genetic architectures, including significant non-additive effects.
• GWKBR software is publicly available for broader application.