Multiview-cooperated graph neural network enables novel multi-omics cancer subtype classification

Area of Science:

• Computational Biology
• Bioinformatics
• Cancer Research

Background:

• Cancer's heterogeneity poses significant public health challenges.
• Multi-omics data integration offers deeper insights into cancer biology and subtyping.
• Existing methods struggle with data scale and analyzing shared/individual feature expressions across omics.

Purpose of the Study:

• To develop an advanced computational model for integrating and analyzing multi-omics data for cancer subtype classification.
• To introduce the Multiview-Cooperated graph neural network (MCgnn) as an effective end-to-end classifier.
• To enhance understanding of cancer biology and identify potential biomarkers through integrated omics analysis.

Main Methods:

• Constructing a similarity network using Mahalanobis distance and density methods.
• Employing stacked graph convolution layers to capture local structural features.
• Utilizing an attention mechanism for fusing complementary information across different omics views.
• Implementing multi-task learning with a cross-omics tensor for integrated feature learning and classification.

Main Results:

• MCgnn demonstrated superior performance in cancer subtype classification compared to existing algorithms on TCGA datasets.
• The model exhibited robust generalization capabilities across multiple cancer types.
• MCgnn successfully identified pivotal biomarkers, offering valuable insights for precision medicine.

Conclusions:

• MCgnn provides an effective framework for integrating multi-omics data for cancer research.
• The developed model advances cancer subtype classification and biomarker discovery.
• This approach holds promise for improving precision medicine strategies in oncology.