Comparison of statistical procedures for estimating polygenic effects using dense genome-wide marker data

Area of Science:

• Quantitative genetics
• Animal breeding
• Statistical genomics

Background:

• Accurate estimation of breeding values is crucial for genetic improvement in livestock.
• Genomic selection utilizes high-density single nucleotide polymorphism (SNP) data to predict breeding values.
• Developing efficient statistical methods for SNP effect estimation is an ongoing research area.

Purpose of the Study:

• To compare the performance of different statistical procedures for estimating SNP effects.
• To evaluate the accuracy of genomic breeding values estimated using ridge regression variants and a mixed model approach.

Main Methods:

• Utilized a simulated dataset from the XII QTL-MAS workshop.
• Compared five statistical procedures, including variants of ridge regression with different shrinkage parameters.
• Assessed correlations between genomic estimated breeding values (GEBVs) and conventional estimated breeding values (EBVs).
• Predicted GEBVs in independent generations and correlated them with true breeding values (TBVs).

Main Results:

• All tested ridge regression methods yielded moderate correlations between GEBVs and EBVs.
• Two selected ridge regression methods showed moderate correlations with TBVs when predicting in later generations.
• The performance of ridge regression procedures was comparable to a simple mixed model method using a ratio of variances.

Conclusions:

• Ridge regression procedures, as applied in this study, did not offer superior accuracy over a basic mixed model approach for genomic breeding value prediction.
• Both methods provided moderate accuracies for predicted genomic breeding values.
• Further research may be needed to optimize statistical methods for genomic selection in complex traits.