DeepOmix: A scalable and interpretable multi-omics deep learning framework and application in cancer survival...

Area of Science:

• Computational Biology and Bioinformatics
• Genomics and Molecular Biology
• Machine Learning in Healthcare

Background:

• Integrating diverse omics data (genomics, transcriptomics, etc.) is crucial for understanding complex diseases but faces challenges due to data heterogeneity and high dimensionality.
• Accurate survival prediction and identification of functional gene modules from multi-omics data are critical unmet needs in precision oncology.

Purpose of the Study:

• To develop a powerful, scalable, and interpretable deep learning framework for multi-omics data integration.
• To enhance cancer survival prediction accuracy and identify biologically relevant functional gene modules.
• To apply the framework to Lower Grade Glioma (LGG) for prognosis prediction and module discovery.

Main Methods:

• Developed DeepOmix, a flexible deep learning framework for non-linear integration of multi-omics data.
• Incorporated user-defined biological information, such as signaling pathways and tissue networks, into the model.
• Validated DeepOmix using benchmark experiments against five state-of-the-art prediction methods.

Main Results:

• DeepOmix demonstrated superior performance compared to five existing cutting-edge prediction methods in survival analysis.
• The framework successfully predicted prognosis for Lower Grade Glioma (LGG) patients.
• Identified specific functional gene modules associated with prognostic outcomes in the LGG case study.

Conclusions:

• DeepOmix provides a robust and interpretable solution for multi-omics data integration in cancer research.
• The framework significantly advances the capability for accurate cancer survival prediction and biomarker discovery.
• DeepOmix holds promise for improving clinical decision-making and personalized treatment strategies in oncology.