TF-DWGNet: a directed weighted graph neural network with tensor fusion for multi-omics cancer subtype classification.
Summary
This paper introduces TF-DWGNet, a novel Graph Neural Network framework developed for multi-omics cancer subtype classification. The model features two key innovations: a supervised tree-based strategy for constructing directed weighted graphs tailored to each omics modality, and a tensor fusion mechanism to efficiently capture unimodal, bimodal, and trimodal interactions. Experiments on three real-world cancer datasets demonstrate that TF-DWGNet consistently outperforms state-of-the-art baselines. Furthermore, it provides valuable interpretable insights through modality-level contribution scores and ranked feature importance.
Analysis
The integration and analysis of multi-omics data are crucial for accurate cancer subtype classification, which is foundational for personalized medicine. TF-DWGNet represents a significant advancement by offering a more robust and interpretable approach to handle the inherent complexity of heterogeneous multi-omics data. Its ability to identify specific modality contributions and feature importance could enhance our understanding of underlying cancer mechanisms, potentially leading to improved diagnostics and more targeted therapeutic strategies.