ML-MAGES enables multivariate genetic association analyses with genes and effect size shrinkage

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) are crucial for identifying genetic variants linked to traits.
• GWAS face challenges including inflated effect estimates and analyzing multiple traits simultaneously.

Purpose of the Study:

• To develop a computationally efficient machine learning method (ML-MAGES) to address inflation in GWAS effect estimates.
• To enable simultaneous investigation of genetic associations across multiple traits.

Main Methods:

• Utilized neural networks for shrinkage of inflated GWAS effect sizes due to variant non-independence.
• Employed variational inference for clustering variant associations across multiple traits.
• Compared neural network shrinkage to regularized regression and fine-mapping.

Main Results:

• Neural network shrinkage demonstrated superior performance in approximating true effect sizes compared to existing methods in simulations.
• The infinite mixture clustering approach effectively distinguished trait-specific, shared, and non-prioritized associations.
• ML-MAGES achieved higher precision and recall in identifying gene-level associations in simulations.
• Application to UK Biobank data identified relevant genes and revealed shared multitrait associations.

Conclusions:

• ML-MAGES offers an efficient and flexible approach for robust genetic association analysis.
• The method successfully addresses key challenges in GWAS, enhancing the identification of genetic architectures.
• Findings suggest potential shared genetic underpinnings across different quantitative and binary traits.