Bayesian divergence-based approach for genomic multitrait ordinal selection

Effective genomic selection for ordinal traits, such as disease resistance scores, is a persistent challenge in plant breeding due to the discrete, ordered nature of these phenotypes. This study presents a novel Bayesian divergence-based framework for multitrait ordinal selection, implemented in the extended Multitrait Parental Selection R package (MPS-R). By leveraging decision-theoretic loss functions, including the Kullback-Leibler (KL) divergence, Bhattacharyya distance, and Hellinger distance, our approach quantifies the distance between candidate distributions and breeder-defined target distributions. Through extensive simulations under 6 scenarios combining different genetic correlation structures and heritability levels, we demonstrate the comparative performance of each loss function. KL divergence consistently yielded superior genetic gains, especially in moderate heritability settings. Additionally, random sampling validation using real wheat disease resistance data confirmed the utility of these methods in practical breeding contexts. The MPS-R package implements this methodology through user-friendly functions tailored for ordinal trait selection in breeding applications. Our results demonstrate that this toolset provides a flexible, robust, and biologically grounded framework to enhance selection efficiency in breeding programs targeting complex, multitrait ordinal phenotypes. A couple of limitations employed by the simulation scheme used on the study are also discussed.