Principal component approach in variance component estimation for international sire evaluation.

Area of Science:

• Animal Breeding and Genetics
• Quantitative Genetics
• Statistical Genetics

Background:

• The global dairy cattle breeding industry requires reliable, internationally comparable breeding values.
• Interbull (International Bull Evaluation Service) was established to provide these international breeding values.
• Estimating parameters for Multiple-Trait Across Country Evaluations (MACE) is complex due to over-parameterized genetic covariance matrices.

Purpose of the Study:

• To compare two principal component approaches for estimating variance components in MACE.
• To evaluate the utility of the bottom-up principal component approach for determining covariance matrix rank.
• To address challenges in estimating parameters for MACE in dairy cattle.

Main Methods:

• Comparison of a direct principal component (PC) REML approach with a bottom-up PC REML approach.
• Utilizing real datasets for variance component estimation in MACE.
• Sequential addition of traits in the bottom-up PC approach to identify significant genetic principal components.

Main Results:

• Both principal component approaches are effective for large, multi-country MACE models.
• These methods can replace current practices of subset analyses for covariance component estimation.
• Appropriate rank selection is crucial; too low a rank causes bias, while too high a rank increases errors and computation time.

Conclusions:

• Both principal component approaches yield accurate estimations and enable parsimonious random regression MACE models.
• The bottom-up PC approach identifies rank without prior knowledge.
• The direct PC approach is faster if the rank is predetermined.