GL-Fusion: A Multi-Omics Integration Method Based on Graph-Level Structure Fusion and Locus-Level Feature Fusion for...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data offers new avenues for cancer subtype classification.
• Graph-based methods model gene associations but struggle with high-dimensional data and complex inter-omic relationships.
• Existing methods often focus on either graph structure or feature-level fusion, limiting comprehensive modeling.

Purpose of the Study:

• To propose GL-Fusion, a novel multi-omics integration method.
• To combine graph-level structure fusion and locus-level feature fusion for enhanced cancer subtype classification.
• To address limitations in existing graph construction and multi-omics fusion techniques.

Main Methods:

• Developed GL-Fusion, integrating multi-omics gene-gene graphs using similarity network fusion.
• Optimized graph structure using structural entropy and protein-protein interaction networks.
• Employed locus-level graph convolutional networks for multi-omics feature integration and fused representation learning.

Main Results:

• GL-Fusion demonstrated superior performance in cancer subtype classification across four datasets (BRCA, HNSC, LGG, THCA).
• Outperformed 12 representative multi-omics cancer subtyping methods.
• Ablation studies and cross-omics association evaluations validated the method's effectiveness.

Conclusions:

• GL-Fusion effectively integrates multi-omics data through combined graph and feature fusion.
• The method enhances cancer subtype classification accuracy.
• GL-Fusion provides a robust framework for leveraging complex multi-omics data in cancer research.