Bayesian genomic-enabled prediction as an inverse problem.

Area of Science:

• Genomics and Quantitative Genetics
• Statistical Genetics in Breeding

Background:

• Genomic prediction is crucial in plant and animal breeding.
• Collinearity arises when the number of molecular markers (p) exceeds sample size (n).
• Existing models struggle with high-dimensional genomic data.

Purpose of the Study:

• To propose novel Bayesian approaches for genomic prediction.
• To enhance prediction accuracy by addressing marker collinearity.
• To reduce computational demands in genomic selection.

Main Methods:

• Utilized a Gaussian linear model for orthogonal transformation of data and marker matrices.
• Developed four distinct prior variance structures for transformed effects.
• Applied data reduction techniques within a Bayesian framework.

Main Results:

• The proposed Bayesian models achieved a 3% average increase in prediction accuracy for maize and wheat data.
• Demonstrated improved prediction accuracy compared to standard Bayesian regression models.
• Showcased reduced computational effort in genomic prediction.

Conclusions:

• The novel Bayesian methods offer a significant improvement in genomic prediction accuracy.
• These approaches effectively handle collinearity in high-dimensional genomic data.
• The methods provide a computationally efficient alternative for breeding applications.