Estimating variance components and predicting breeding values for eventing disciplines and grades in sport horses.

This research attempts to improve the prediction of breeding values in sport horses by devising different methods to calculate large (co)variance matrices using data from Great Britain’s eventing competitions.

Background

• The study bases itself on the GB’s eventing competitions, which comprise three disciplines each split into four grades, a total of 12 discipline-grade traits.
• The creation of tools to estimate covariance matrices for these 12 traits is crucial as they contribute to valuable insights regarding variance components and breeding value predictions.

Data Collection and Usage

• The researchers used data from 1999 to 2008, converting penalty points to normal scores to standardize measurements.
• A model was used to estimate fixed effects like gender, age, and class, and random effects of sire, horse, and rider.

Methods for Estimating Covariance Matrices

• Three methods were used to estimate covariance matrices. ‘Method 1’ employed Gibbs sampling and data augmentation and imputation.
• ‘Methods 2a and 2b’ were combined sub-matrices from bivariate analyses. Method 2a took samples from a multivariate Normal distribution defined by each bivariate analysis’s covariance matrix and then analyzed these data in a 12-trait multivariate analysis.
• Method 2b replaced negative eigenvalues within the matrix with positive values to achieve a positive, definite covariance matrix.
• A formal comparison between these methods wasn’t possible, but the estimates produced by all methods were comparatively reasonable.

Outcomes of Estimation Methods

• In terms of computational requirements, Method 1 required significantly less compared to Methods 2a and 2b.
• The heritability estimates varied across methods: Method 2a heritability estimates for dressage, show jumping and cross-country fell within 7.2% – 9.0%, 8.9% – 16.2% and 1.3% – 1.4%, respectively, while Method 1 yielded higher heritability estimates for the advanced grades.
• Regardless of the estimation method, genetic correlations between grades for dressage and show jumping were significant and highly positive. For cross-country, significant correlations were only identified between novice and pre-novice using Method 2a, but all estimates in Method 1 were significant.
• All three methods produced positive definite 12 × 12 covariance matrices, making them suitable for the prediction of breeding values.

Significance

• This research contributes to better predict breeding values by devising new methods to estimate covariance matrices using competition data.
• Method 1 proved to be computationally more efficient and can perform a true multivariate analysis, distinguishing it from the others used.