Penalized co-inertia analysis with applications to -omics data.

Area of Science:

• Multivariate statistics
• Bioinformatics
• Genomics

Background:

• Co-inertia analysis (CIA) integrates two datasets but struggles with interpretation due to non-sparse loading vectors in high-dimensional omics data.
• Existing methods like penalized least squares (PLS) offer sparse solutions, but CIA lacks similar advancements for omics integration.

Purpose of the Study:

• To develop novel penalized co-inertia analysis (CIA) methods for improved omics data integration and interpretation.
• To introduce a CIA approach that incorporates biological network information for enhanced biological relevance.

Main Methods:

• Proposed a novel CIA method using l1-penalization to induce sparsity in loading vector estimators, enabling simultaneous model fitting and variable selection.
• Developed a second CIA method integrating functional genomics network information with sparsity penalties for biologically interpretable results.

Main Results:

• Extensive simulations show proposed penalized CIA methods outperform existing methods in feature selection and loading vector recovery.
• Applied to NCI-60 cancer cell line data, revealing key variables and biologically consistent findings for cancer research.

Conclusions:

• The novel penalized CIA methods offer a powerful and interpretable approach for integrative omics data analysis.
• These methods facilitate the discovery of biologically meaningful insights from complex, high-dimensional datasets.