L2,1-norm regularized multivariate regression model with applications to genomic prediction

Area of Science:

• Agricultural Science
• Genetics
• Bioinformatics

Background:

• Genomic selection (GS) accelerates crop breeding by leveraging genomic data.
• Multi-trait GS can improve prediction accuracy for traits with low heritability.
• Current GS methods often lack mechanistic understanding of single nucleotide polymorphism (SNP) contributions.

Purpose of the Study:

• To introduce a novel L2,1-norm regularized multivariate regression model for multi-trait genomic selection.
• To develop an efficient algorithm (L2,1-joint) for this model.
• To enable variable selection for dissecting genetic architecture and identifying key genetic regulators.

Main Methods:

• Proposed a L2,1-norm regularized multivariate regression model.
• Devised a fast iterative optimization algorithm named L2,1-joint.
• Applied the model to multi-trait GS, considering relationships between individuals and handling a large number of SNPs relative to individuals.

Main Results:

• The L2,1-joint model demonstrated superior performance compared to existing state-of-the-art approaches in comparative analyses.
• Variable selection capability was achieved, facilitating the identification of potential master regulators.
• The model's effectiveness was validated using diversity panels from Brassica napus, wheat, and Arabidopsis thaliana.

Conclusions:

• The L2,1-joint model offers a powerful approach for multi-trait genomic selection, enhancing prediction accuracy and providing insights into genetic architecture.
• This method facilitates mechanistic understanding by enabling variable selection of SNPs.
• The R implementation is freely available, promoting its adoption in plant breeding research.