Principal component and factor analytic models in international sire evaluation.

Area of Science:

• Quantitative genetics
• Animal breeding
• Statistical modeling

Background:

• International dairy cattle breeding programs face challenges due to diverse national genetic evaluation models.
• High genetic correlations between countries necessitate parsimonious models to avoid over-parameterization.
• Existing multi-trait models can be computationally intensive and lead to increased sampling variances.

Purpose of the Study:

• To evaluate the utility of Principal Component (PC) and Factor Analytic (FA) models for Multiple-Trait Across Country Evaluation (MACE).
• To assess the accuracy of variance component estimation and breeding value prediction using PC and FA models.
• To compare the performance of PC and FA models against traditional full-rank models in terms of accuracy and computational efficiency.

Main Methods:

• Applied Principal Component (PC) and Factor Analytic (FA) models to Holstein bull protein yield evaluations from 25 countries.
• Determined the optimal number of components/factors explaining genetic variation.
• Compared genetic parameter estimates, breeding value predictions, and estimation times with full-rank models.

Main Results:

• Optimal PC (19) and FA (9) models explained genetic variation effectively, yielding highly similar genetic parameter estimates.
• Estimates from PC and FA models closely agreed with full-rank models and Interbull's official evaluations.
• Optimal model fitting minimized estimation time; over-fitting increased computation time and standard errors without impacting correlations, while under-fitting affected bull rankings.

Conclusions:

• PC and FA models provide accurate and similar genetic parameter estimates when optimally fitted.
• Optimal model selection is crucial for balancing computational efficiency and accurate breeding value prediction in MACE.
• Under-fitting PC/FA models can lead to biased bull rankings across countries, highlighting the importance of appropriate model complexity.