Genomic Prediction from Multiple-Trait Bayesian Regression Methods Using Mixture Priors.

Area of Science:

• Quantitative genetics
• Statistical genomics
• Bioinformatics

Background:

• Bayesian multiple-regression methods with mixture priors are standard for genomic prediction.
• Existing multi-trait methods assume loci affect all or no traits, which is biologically unrealistic.
• Previous single-trait methods (BayesB, BayesC, BayesCπ) show improved prediction accuracy.

Purpose of the Study:

• To develop and implement generalized multi-trait BayesCπ and BayesB methods.
• To allow loci to affect any combination of traits, moving beyond restrictive assumptions.
• To compare the performance of new methods against single-trait and restrictive multi-trait methods.

Main Methods:

• Developed generalized multi-trait BayesCπ and BayesB models with broader mixture priors.
• Implemented methods allowing loci to influence any subset of traits.
• Compared new methods with single-trait and restrictive multi-trait approaches using simulated and real data.

Main Results:

• Real data analysis showed higher prediction accuracies for both new broad-based and restrictive multi-trait methods.
• Broad-based and restrictive multi-trait methods yielded similar prediction accuracies on real data.
• Simulated data analysis indicated superior performance of general multi-trait methods over the restrictive approach for intermediate training population sizes.

Conclusions:

• The developed broad-based multi-trait Bayesian methods offer a more biologically realistic framework for genomic prediction.
• These methods provide comparable or improved prediction accuracies over existing approaches.
• The JWAS software tool is available for implementing these advanced genomic prediction analyses.