Combining Sparse Group Lasso and Linear Mixed Model Improves Power to Detect Genetic Variants Underlying Quantitative...

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genome-Wide Association Studies (GWAS) traditionally analyze single nucleotide polymorphisms (SNPs) individually, limiting comprehensive analysis of complex traits.
• Complex traits are polygenic, necessitating methods that consider groups of SNPs simultaneously and account for confounding factors like population structure.

Purpose of the Study:

• To develop a novel statistical method, SGL-LMM, integrating Sparse Group Lasso (SGL) and Linear Mixed Models (LMM) for multivariate quantitative trait association analysis.
• To improve the power of detecting genetic associations and the accuracy of quantitative trait prediction in GWAS.

Main Methods:

• SGL-LMM combines LMM for controlling confounding effects and SGL for maintaining sparsity in multivariate regression.
• The method involves estimating random effects using LMM and fixed effects using SGL regularization.
• Efficient algorithms for hyperparameter tuning and feature selection via stability selection are presented.

Main Results:

• SGL-LMM effectively controls for confounders while enforcing sparse solutions.
• The approach naturally incorporates prior biological information into the model's group structure.
• Simulated and real data analyses demonstrate superior performance of SGL-LMM over existing methods in association power and prediction accuracy.

Conclusions:

• SGL-LMM offers a robust framework for multivariate association analysis in GWAS.
• The method enhances the understanding of complex trait genetics by simultaneously considering SNP groups and confounding factors.
• SGL-LMM represents a significant advancement in statistical genetics for complex trait dissection.