Genomic Selection for Milk Yield and Milk Composition Traits in Dairy Goats Using Machine Learning and...

Area of Science:

• Animal genetics and breeding
• Genomic prediction methodologies
• Dairy goat production

Background:

• Genomic selection (GS) accelerates genetic gain in dairy goats but is sensitive to model choice, marker density, and trait architecture.
• Accurate genomic prediction is crucial for efficient dairy goat breeding programs.

Purpose of the Study:

• To evaluate and compare different statistical models and genotyping strategies for genomic prediction of milk yield and composition in Chinese dairy goats.
• To identify optimal approaches for enhancing prediction accuracy in dairy goat breeding.

Main Methods:

• Genomic data from 1034 Xinong Saanen and Saanen dairy goats were generated using low-coverage whole-genome sequencing (lcWGS) and imputed.
• A 25K single-nucleotide polymorphism (SNP) chip dataset was also used.
• Conventional genomic best linear unbiased prediction (GBLUP), Bayesian regression (BayesB), and machine learning (XGBoost) models were compared via 10-fold cross-validation.

Main Results:

• Bayesian models, particularly BayesB, improved milk fat percentage (MFP) prediction by 12.9% over GBLUP.
• Extreme gradient boosting (XGBoost) enhanced prediction accuracy for milk yield (MY), MFP, and milk protein percentage (MPP) by 14.3%, 17.9%, and 18.5%, respectively, compared to GBLUP.
• Incorporating genome-wide association study (GWAS) and selection signature priors further boosted prediction accuracy, especially for milk composition traits.

Conclusions:

• Advanced models (Bayesian, XGBoost) and genotyping strategies significantly improve genomic prediction accuracy in dairy goats.
• Tailoring models and incorporating prior biological information to specific traits is key to optimizing genomic prediction for dairy goat breeding.