Sparse matrix factorization robust to sample sharing across GWASs reveals interpretable genetic components.

Area of Science:

• Genetics
• Bioinformatics
• Statistical genetics

Background:

• Complex traits exhibit extensive genetic pleiotropy, meaning variants influence multiple phenotypes.
• Multi-phenotype analyses are crucial for identifying shared and specific genetic factors.
• Existing matrix factorization methods for GWASs suffer from confounding due to sample sharing and produce dense, hard-to-interpret factors.

Purpose of the Study:

• To introduce GLEANR (GWAS latent embeddings accounting for noise and regularization), a novel matrix factorization method for detecting sparse genetic factors from GWAS summary statistics.
• To address limitations of previous methods, including confounding from sample overlap and difficulty in biological interpretation.
• To improve the discovery and replication of interpretable genetic factors.

Main Methods:

• Developed GLEANR, a matrix factorization approach incorporating noise and regularization.
• Designed GLEANR to account for sample sharing between GWASs.
• Implemented regularization to estimate a data-driven number of sparse, interpretable factors.

Main Results:

• Applied GLEANR to 137 UK Biobank GWASs, identifying 58 interpretable genetic factors.
• Demonstrated GLEANR's robustness to confounding from shared samples, improving factor replication.
• Characterized identified factors by negative selection signatures, polygenicity, and enrichment for disease, cell type, and pathways.

Conclusions:

• GLEANR is a powerful tool for dissecting the genetic architecture of complex traits using GWAS summary statistics.
• The method effectively identifies sparse, biologically interpretable genetic factors, including trait-specific and trait-shared pathways.
• GLEANR facilitates deeper understanding of pleiotropy and the genetic basis of complex diseases.