Genomic prediction for numerically small breeds, using models with pre-selected and differentially weighted markers

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Genomic Prediction

Background:

• Genomic prediction (GP) accuracy is often limited in small dairy breeds due to small reference populations.
• A novel multi-breed multiple genomic relationship matrices (GRM) GP model (MBMG) was developed to address this limitation.

Purpose of the Study:

• To evaluate the effectiveness of the MBMG model in improving GP accuracy for stature in numerically small breeds.
• To compare the MBMG model against single GRM and within/across-breed models.

Main Methods:

• Utilized genotype and phenotype data from Jersey and Holstein bulls with deregressed proofs for stature.
• Implemented a multi-breed bivariate GREML model fitting either a single GRM (MBSG) or two distinct GRMs (MBMG) using pre-selected and remaining markers.
• Compared prediction accuracies for Jersey individuals using combined (Holstein and Jersey) versus single-breed reference populations.

Main Results:

• The MBMG model consistently yielded higher prediction accuracies for stature compared to MBSG and within/across-breed models.
• MBMG achieved an accuracy of 0.49 using pre-selected and unselected markers in separate GRMs, outperforming MBSG (0.43).
• Simulation studies confirmed MBMG's superiority, showing an average 23% improvement over MBSG.

Conclusions:

• The MBMG model effectively enhances GP accuracy in small breeds by incorporating data from larger breeds.
• MBMG's improved performance stems from its ability to utilize pre-selected markers, explain residual genetic variance, and weight breed information by genetic correlation.