Multi-Trait Bayesian Models Enhance the Accuracy of Genomic Prediction in Multi-Breed Reference Populations

Background

Combining multiple breeds in a genomic reference population can increase the reference size, but simple pooling may fail when breeds differ in genetic background and marker-effect covariance structure.

Results

This study proposed mbBayesAB, a multibreed genomic prediction model that treats the same trait in different breeds as genetically correlated but distinct traits. It partitions chromosomes into blocks and fits heterogeneous genetic covariance across these blocks. The study also evaluated prior settings for random-effect covariance matrices and compared mbBayesAB with within-breed genomic prediction and commonly used multibreed models.

Conclusion

Modeling heterogeneous covariance in multibreed genomic prediction can improve prediction accuracy, but appropriate covariance priors and hyperparameters are critical for stable performance.