Robust graph structure learning to improve multi-omics cancer subtype classification.

Area of Science:

• Computational Biology and Bioinformatics
• Cancer Genomics
• Precision Medicine

Background:

• Accurate cancer subtyping using multi-omics data is crucial for precision medicine but challenging due to complex data integration.
• Integrating intra-omics and inter-omics information alongside sample networks presents a significant hurdle in multi-omics data analysis.

Purpose of the Study:

• To develop an advanced computational model for precise multi-omics cancer subtyping.
• To effectively integrate diverse omics data, patient network information, and survival data for improved classification.

Main Methods:

• Introduced the Feature and Graph Structure-Learning Integrated Graph Convolutional Network (FaGGCN) model.
• Employed convolutional autoencoders for latent feature extraction and survival analysis for feature selection.
• Utilized a graph autoencoder for inter-omics similarity fusion and a graph convolutional network for patient classification.

Main Results:

• The FaGGCN model demonstrated robust performance across eight cancer datasets with four omics modalities.
• Achieved improved cancer patient classification and exploratory survival prediction.
• Survival, sensitivity, and differential gene expression analyses confirmed model interpretability and biomarker identification capabilities.

Conclusions:

• The FaGGCN model offers a competitive and effective approach for multi-omics cancer subtyping.
• The model's ability to integrate complex data enhances classification accuracy and survival prediction.
• Identified biomarkers show potential for clinical research applications.