Improving the identification of cancer driver modules using deep features learned from multi-omics data.

Area of Science:

• Computational biology
• Cancer genomics
• Systems biology

Background:

• Identifying cancer driver modules is essential for understanding cancer mechanisms.
• Abundant cancer omics data offers opportunities but existing methods struggle with integration.
• Limitations exist in current computational methods for effectively learning omics features for driver module identification.

Purpose of the Study:

• To develop an integrated framework for accurate cancer driver module identification.
• To overcome limitations of existing methods in integrating diverse cancer omics data.
• To improve the learning of informative omics features for driver module discovery.

Main Methods:

• Integrating protein-protein interaction networks, transcriptional regulatory networks, gene expression, and mutation data.
• Developing methods to learn deep features of functional gene connectivity across omics data.
• Constructing an integrated gene functional coherence network.
• Applying a two-step module mining method to identify driver modules.

Main Results:

• The proposed framework successfully integrates multiple omics data types.
• Deep features of functional connectivity were learned for each omics data.
• An integrated gene functional coherence network was constructed.
• A two-step module mining method efficiently identified cancer driver modules.
• Systematic experiments in three cancer types demonstrated superior performance over existing methods.
• Identified driver modules showed association with clinical survival phenotypes.

Conclusions:

• The novel integrated framework accurately identifies cancer driver modules.
• The method effectively leverages diverse omics data for enhanced driver module discovery.
• The findings contribute to a better understanding of cancer progression and potential therapeutic targets.