Bayesian methods for estimating GEBVs of threshold traits.

Area of Science:

• Quantitative genetics
• Animal breeding
• Statistical genetics

Background:

• Genomic selection (GS) relies on estimating genomic breeding values (GEBVs).
• Existing methods primarily focus on continuous traits, leaving a gap for threshold traits.
• Threshold traits, common in animal and plant breeding, require specialized modeling.

Purpose of the Study:

• To extend existing Bayesian methods (BayesA, BayesB, BayesCπ) for GEBV estimation to threshold traits.
• To introduce and evaluate novel threshold models: BayesTA, BayesTB, and BayesTCπ.
• To assess the accuracy and factors influencing GEBV prediction for threshold traits.

Main Methods:

• Development of three Bayesian threshold models (BayesTA, BayesTB, BayesTCπ) within the GS framework.
• Derivation of Markov Chain Monte Carlo (MCMC) computational procedures for the BayesT methods.
• Simulation studies to evaluate prediction accuracy and compare performance against standard Bayesian methods.

Main Results:

• The proposed BayesT methods generally outperform standard Bayesian methods for threshold traits, especially with fewer phenotypic categories.
• Significant accuracy improvements were observed: 30.4% (BayesTA), 2.4% (BayesTB), and 5.7% (BayesTCπ) in a standard scenario.
• BayesTB and BayesTCπ demonstrated comparable and superior performance over BayesTA across most simulated scenarios.

Conclusions:

• The threshold model is well-suited for predicting GEBVs of threshold traits.
• BayesTCπ emerges as the recommended method for achieving higher accuracy in genomic selection for threshold traits.
• The developed methods offer valuable tools for genetic improvement programs dealing with categorical trait data.