Improving multibreed genomic prediction for breeds with small populations by modeling heterogeneous genetic (co)variance blockwise accounting for linkage disequilibrium

Background

Multibreed genomic prediction is important for improving prediction accuracy in breeds with limited population size and limited phenotypic records. Previous work has shown that modeling heterogeneous genetic covariance across genomic blocks can improve joint prediction, but the strategy for defining these blocks had not been systematically evaluated.

Results

This study proposed mbBayesABLD, a Bayesian multibreed genomic prediction model that partitions each chromosome into non-overlapping linkage disequilibrium blocks and models marker effects with block-specific parameters. Using simulation data and real pig and bean datasets, mbBayesABLD outperformed conventional multibreed genomic prediction models such as GBLUP and BayesR. In the pig intramuscular fat score trait, mbBayesABLD improved the prediction accuracy of the small Landrace population by 15.6% compared with GBLUP without heterogeneous covariance modeling.

Conclusion

The study provides a practical LD-block-based method for multibreed genomic prediction and shows how between-breed genetic diversity can be used to improve genomic prediction in small populations.