Application of multi-trait Bayesian decision theory for parental genomic selection

Area of Science:

• Quantitative genetics
• Plant breeding
• Statistical genomics

Background:

• Multi-trait selection is crucial for improving genetic stocks, as economic value depends on numerous traits.
• Antagonistically correlated traits pose challenges in selection, where improving one trait can negatively impact another.
• Genomic selection (GS) and advanced statistical tools are driving a focus on multi-trait selection strategies.

Purpose of the Study:

• To apply Bayesian decision theory (BDT) for complex multi-trait parental selection in wheat breeding.
• To evaluate the effectiveness of three multivariate loss functions (KL, Energy Score, MALF) within the BDT framework.
• To assess how BDT aids breeders in making selection decisions considering trait interdependencies and uncertainty.

Main Methods:

• Application of Bayesian decision theory (BDT) to multi-trait parental selection.
• Utilized three distinct multivariate loss functions: Kullback-Leibler (KL), Energy Score, and Multivariate Asymmetric Loss (MALF).
• Tested the BDT approach on two extensive real wheat datasets.

Main Results:

• High genomic estimated breeding values (GEBV) for specific traits did not consistently correlate with low posterior expected loss (PEL).
• The Kullback-Leibler (KL) loss function treated all traits, including grain yield, with equal importance.
• The Energy Score and Multivariate Asymmetric Loss (MALF) demonstrated superior performance for traits other than grain yield.

Conclusions:

• BDT provides a comprehensive approach to multi-trait parental selection by integrating heritability, selection response, and trait correlations.
• The choice of loss function within BDT influences the prioritization of different traits during selection.
• BDT enhances breeding decisions by incorporating the level of uncertainty alongside GEBV, offering a more nuanced selection process.