Bayesian prediction of breeding values for multivariate binary and continuous traits in simulated horse populations using threshold-linear models with Gibbs sampling.

The research article discusses the utilization of simulated horse data to predict breeding values for categorical and continuous traits using a mixed linear-threshold animal model. The primary goal is to aid in achieving the maximum reduction of binary traits in a horse population through the use of phenotype and genetic marker data.

Methodology

• The researchers used simulated data as a testing ground for the analysis. The simulations entailed creating a seven-generation lineage of 40,000 animals in each round. The data generated covered both additive genetic values and residual fixed effects for one continuous trait and liabilities for four binary traits.
• Additionally, for one of the binary traits, effects of the quantitative trait locus (QTL) and genetic markers were simulated. Three different conditions were tested: high recombination rate with high quality genetic markers (r = 0.01 and PIC = 0.90), no recombination rate with high quality genetic markers (r = 0.00 and PIC = 0.90), and no recombination rate with relatively less informative genetic markers (r = 0.00 and PIC = 0.70).
• For each scenario, 10 replicates were sampled, creating six different data sets per replicate. These datasets differed based on the quantity, distribution of animals, and availability of genetic marker information.

Data Analysis

• The breeding values were predicted using a specialized method known as Gibbs sampling. This was executed within the Bayesian mixed linear-threshold animal model. This method allows for the proper prediction of breeding values even when the residual covariances are fixed to zero.
• The derived relative breeding values were used to investigate the possible outcomes of single- and multi-trait selection.

Results and Conclusion

• The correlation between true and predicted breeding values for continuous traits was found to be between 0.89 and 0.94, and for binary traits, it ranged from 0.39 to 0.77.
• The accuracy in identifying sires with average, favorable, or unfavorable genetic value was about 81% to 86% for the continuous trait and 57% to 74% for the binary traits.
• The expected decrease in the prevalence of the QTL trait after single-trait selection was between 9% and 17%, and after multi-trait selection, it was between 3% and 12%.
• The use of phenotype and genotype data combined for prediction was found to be more effective compared to using phenotype data alone.
• Consequently, the study suggests the use of phenotype and highly informative genetic markers data in a mixed linear-threshold animal model via Gibbs sampling for predicting breeding values. This method will potentially lead to the maximum reduction in the prevalences of binary traits.