Sparse reduced-rank regression for integrating omics data.

Area of Science:

• Biostatistics
• Genomics
• Metabolomics
• Complex Disease Research

Background:

• Assessing associations between multiple omics data (genomics, metabolomics) for complex disease biomarkers is challenging.
• Univariate regression is common but requires large sample sizes and ignores within- and across-data type correlations.
• Existing methods struggle with the complexity of multi-omics data integration.

Purpose of the Study:

• To develop a novel statistical framework for integrating multi-omics data.
• To identify multiple relevant predictors simultaneously associated with multiple responses.
• To improve biomarker discovery for complex diseases.

Main Methods:

• Utilized a multivariate linear regression model for multi-omics data.
• Assumed a row-sparse and low-rank coefficient matrix.
• Proposed a group Dantzig type formulation for coefficient matrix estimation.

Main Results:

• The proposed method demonstrated competitive performance in estimation, prediction, and variable selection via simulations.
• Successfully integrated genomics and metabolomics data to identify predictive genetic variants for ASCVD.
• Identified genetic variants that enhance ASCVD prediction beyond established risk factors.

Conclusions:

• The novel method offers a robust approach for multi-omics data integration and biomarker discovery.
• The identified genetic variants show potential for improving ASCVD risk prediction.
• The findings suggest the utility of identified genetic variants in explaining metabolite-ASCVD associations.