TF-DWGNet: A Directed Weighted Graph Neural Network with Tensor Fusion for Multi-Omics Cancer Subtype Classification

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration is crucial for cancer subtype classification but faces challenges due to data heterogeneity and complex dependencies.
• Existing graph neural network (GNN) methods often use predefined, undirected graphs, limiting their ability to capture directional interactions and feature importance.
• Limited interpretability in current GNN models hinders biological insight discovery.

Purpose of the Study:

• To introduce TF-DWGNet, a novel GNN framework for multi-omics integration and cancer subtype classification.
• To address limitations in existing GNNs by incorporating directed, weighted graph construction and tensor fusion.
• To enhance model interpretability at both modality and feature levels.

Main Methods:

• Developed a supervised tree-based strategy for constructing directed, weighted graphs tailored to each omics modality.
• Implemented a tensor fusion mechanism with low-rank decomposition to efficiently capture unimodal, bimodal, and trimodal interactions.
• Evaluated TF-DWGNet on three real-world cancer datasets.

Main Results:

• TF-DWGNet consistently outperformed state-of-the-art methods across multiple metrics and statistical tests.
• The model demonstrated superior performance in multiclass cancer subtype classification.
• Achieved biologically meaningful insights through modality-level contribution scores and ranked feature importance.

Conclusions:

• TF-DWGNet offers an effective and interpretable solution for multi-omics data integration in cancer research.
• The framework successfully captures complex dependencies and interaction strengths within and between omics data.
• Provides a promising tool for advancing cancer subtype classification and understanding.