A comparison of statistical methods for genomic selection in a mice population

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Bioinformatics

Background:

• High-density SNP panels enable marker-assisted selection for predicting genetic merit.
• Marker effect estimation is challenging due to more markers than phenotypes.
• Genomic selection is a powerful tool in animal breeding.

Purpose of the Study:

• To compare the predictive performance of ten statistical methods in genomic selection.
• To analyze genomic selection data from a heterogeneous stock mice population.
• To identify optimal methods for predicting genetic merit.

Main Methods:

• Comparison of ten different statistical methods for genomic selection.
• Analysis of five traits in a heterogeneous stock mice population.
• Evaluation of prediction accuracy, bias, and inflation.

Main Results:

• Within-family predictions were more accurate than across-family predictions, with trait-specific variations.
• Kernel methods (RKHS, SVR) excelled for weight and growth traits.
• Variable selection methods (LASSO, Random Forest) were superior for immune cell traits (%CD8+, CD4+/CD8+).
• RKHS, SVR, RR_GBLUP, and Random Forest showed improved bias and inflation control.

Conclusions:

• Statistical methods showed similar predictive abilities but varied by trait.
• Variable selection methods are effective for traits influenced by fewer quantitative trait loci (QTL).
• RR_GBLUP, RKHS, and SVR are recommended for genomic selection applications based on performance and computational efficiency.