Multi-view multi-level contrastive graph convolutional network for cancer subtyping on multi-omics data.

Area of Science:

• Computational biology
• Bioinformatics
• Cancer research

Background:

• Cancer is a complex disease with diverse subtypes, necessitating accurate classification for effective treatment and prognosis.
• Current multi-omics integration methods often overlook the interplay between consensus and individual information within omics data.
• Existing approaches may not fully leverage the rich relational data present in biological networks.

Purpose of the Study:

• To develop a novel, fusion-free method for accurate cancer subtyping using multi-omics data.
• To address the limitations of existing methods by integrating multi-level features and contrastive learning.
• To improve the understanding of cancer heterogeneity through advanced computational approaches.

Main Methods:

• Proposed a multi-view multi-level contrastive graph convolutional network (M$^{2}$CGCN) for cancer subtyping.
• M$^{2}$CGCN learns both low-level features (intrinsic omics information via reconstruction) and high-level features (cancer subtyping via contrastive learning).
• Employed a fusion-free approach to integrate information across multiple omics levels and views.

Main Results:

• M$^{2}$CGCN demonstrated comparable or superior performance to state-of-the-art methods on 34 multi-omics cancer datasets.
• The method effectively captures intrinsic information within each omics data through reconstruction.
• High-level features derived via contrastive learning significantly contribute to accurate cancer subtyping.

Conclusions:

• The proposed M$^{2}$CGCN method offers a powerful and effective approach for cancer subtyping using multi-omics data.
• This fusion-free strategy enhances the integration of diverse biological data for improved clinical relevance.
• M$^{2}$CGCN provides a promising computational framework for advancing precision oncology.