Estimation of genetic parameters and prediction of breeding values for multivariate threshold and continuous data in a simulated horse population using Gibbs sampling and residual maximum likelihood.

This research study used simulated horse data to compare different methods of genetic parameter estimation and prediction of breeding values. The methods compared included linear animal models and mixed linear-threshold animal models. The results suggest that the consideration of genetic marker information can lead to overestimation of the heritability of the trait. Additionally, it was found that the accuracy of estimates and predictions increases with the number of animals used in the analysis.

Explanation of the Research Paper

• The research study made use of simulated horse population data to assess the performance of different models in performing genetic parameter estimation and prediction of breeding values (BVs). The models in question were the linear animal models and the mixed linear-threshold animal models that utilized residual maximum likelihood (REML) and best linear unbiased prediction (BLUP).
• The simulation included various factors such as additive genetic values, residuals, fixed effects for a continuous trait, liabilities of four binary traits, quantitative trait locus (QTL) effects, and genetic markers having varied rates of recombination and different levels of polymorphism information content for one of the liabilities.
• The studies varied in the number of animals having trait records and the availability of genetic marker information. The research found that using genetic marker information in the model led to overestimation of the heritability of the trait associated with the QTL.
• When information on a sizable number of animals (10,000 or 5,000) was utilized for the analysis, the bias in the heritabilities and additive genetic correlations was significantly smaller. Additionally, the correlation between the true and predicted BVs was consistently higher.
• The researchers found that in terms of identifying genetically superior or inferior animals, particularly for those traits having moderate heritability, the Gibbs sampling (GS) model outperformed the REML/BLUP model.
• Notably, when the dataset had information on only 1,000 animals, neither GS nor REML/BLUP was able to produce genetic parameter estimates having relative bias <=25% or a BV correlation exceeding 50% for all traits.
• The study concluded that when making selection decisions for binary traits, it may be more prudent to base them on the estimations from the GS model rather than the REML/BLUP breeding values.