Multi-omics network reconstruction with collaborative graphical lasso.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Multi-omics data integration is crucial for a comprehensive understanding of biological processes.
• Reconstructing multi-omics networks is a promising integration approach, but effective strategies are lacking.
• Existing methods struggle to fully leverage the complementary information across different molecular layers.

Purpose of the Study:

• To introduce a novel method, collaborative graphical lasso, for improved multi-omics data integration and network inference.
• To enhance the reconstruction of multi-omics networks by harmonizing contributions from different omics layers.
• To develop a robust model selection procedure for multi-dimensional hyperparameter tuning in this framework.

Main Methods:

• Collaborative graphical lasso extends graphical lasso with a collaborative penalty term for harmonized omics layer contributions.
• A dual regularization scheme is employed to control sparsity within and between omics layers.
• The XStARS stability-based criterion is proposed for multi-dimensional hyperparameter tuning.

Main Results:

• Simulations demonstrate the effectiveness of collaborative graphical lasso and the XStARS model selection procedure.
• Application to public multi-omics data successfully recovered known biological interactions.
• The method identified novel, biologically coherent connections, showcasing its potential for new discoveries.

Conclusions:

• Collaborative graphical lasso provides a powerful framework for multi-omics data integration and network inference.
• The proposed XStARS criterion effectively addresses model selection challenges in this complex setting.
• This approach advances the field by enabling more cohesive integration of multi-omics information for biological discovery.