DeepMoIC: multi-omics data integration via deep graph convolutional networks for cancer subtype classification.

Area of Science:

• Computational Biology and Bioinformatics
• Genomics and Multi-omics Data Analysis
• Cancer Research and Precision Medicine

Background:

• Precise cancer subtype classification is crucial for effective prognosis and treatment strategies.
• Multi-omics studies offer powerful insights into cancer complexity but face challenges due to data heterogeneity (variations in types, scales, distributions).
• Existing methods struggle to extract intact representations from heterogeneous multi-omics data, leading to inaccuracies in analysis.

Purpose of the Study:

• To develop a novel framework, DeepMoIC, to address the challenges of analyzing heterogeneous multi-omics data for cancer subtype classification.
• To effectively extract compact and high-order representations from diverse omics data.
• To improve the accuracy and reliability of cancer subtype classification using multi-omics information.

Main Methods:

• Developed DeepMoIC, a novel framework based on deep Graph Convolutional Network (GCN).
• Utilized autoencoder modules for extracting compact data representations.
• Incorporated a patient similarity network using the similarity network fusion algorithm and employed residual connection and identity mapping strategies within the Deep GCN module to handle non-Euclidean data and explore high-order omics information.

Main Results:

• DeepMoIC successfully extracts higher-order representations from complex multi-omics data.
• The proposed approach demonstrates superior performance compared to state-of-the-art models.
• Consistent outperformance was observed on a pan-cancer dataset and three distinct cancer subtype datasets.

Conclusions:

• Deep GCN shows significant promise for supervised multi-omics feature learning, providing valuable insights for precision medicine in cancer.
• DeepMoIC offers a robust solution for handling complex multi-omics data, enabling reliable cancer subtype classification.
• The framework has the potential to become an important tool in advancing cancer research and clinical applications.