Genomic selection for slaughter age in pigs using the Cox frailty model

Area of Science:

• Animal Genetics and Breeding
• Quantitative Genetics
• Statistical Genomics

Background:

• Genomic selection (GS) is crucial for improving livestock traits.
• Accurate prediction of genomic breeding values (GBVs) relies on appropriate statistical models.
• Survival traits, often exhibiting censored data, pose unique challenges for GS.

Purpose of the Study:

• To compare the performance of a linear mixed model and a Cox survival model for genomic selection.
• To evaluate the impact of marker selection on prediction accuracy.
• To identify the most suitable model for analyzing survival traits in pig populations.

Main Methods:

• Utilized an F2 pig population with 238 markers (237 SNPs and the halothane gene).
• Response variable: time to culling; covariates: marker genotypes.
• Compared linear mixed models with Cox survival models (including a Cox frailty model with a normal random effect, S1) for GBV prediction and marker effect estimation.

Main Results:

• Both models showed agreement for uncensored and normally distributed data.
• The Cox model with a normal random effect (S1) was more appropriate for censored data.
• Marker selection enhanced the correlation between predicted GBVs and corrected phenotypic values, with 120 markers optimizing predictive ability.

Conclusions:

• The Cox survival model, particularly with a normal random effect (S1), is superior for genomic selection involving censored survival data in pigs.
• Model S1 effectively accounts for latent variables and censored observations, outperforming standard linear mixed models in such scenarios.
• Strategic marker selection significantly improves the predictive accuracy of genomic breeding values for survival traits.