Structure-Aware Consensus Representation Learning With Dual-Channel Attention for Multi-Omics Cancer Subtype...

Area of Science:

• Computational Biology
• Bioinformatics
• Cancer Research

Background:

• Cancer heterogeneity and subtypes present challenges for diagnosis and treatment.
• Multi-omics clustering integrates diverse biological data for cancer subtype identification.
• Existing methods often overlook omics-specific features and decouple representation learning from clustering.

Purpose of the Study:

• To develop a novel multi-omics clustering method for improved cancer subtype identification.
• To address limitations of existing methods by capturing both unique and common omics features.
• To enhance the joint optimization of feature representations and clustering for better performance.

Main Methods:

• Proposed Structure-Aware Consensus Representation Learning with Dual-Channel Attention (SACR-DCA).
• Implemented a dual-channel attention framework for fusing omics-specific and shared information.
• Utilized structure-aware learning and Cauchy-Schwarz divergence for consensus representation enhancement and clustering adaptability.

Main Results:

• SACR-DCA effectively captures both omics-specific and shared data characteristics.
• The method demonstrates superior performance compared to existing approaches on ten real-world datasets.
• Joint optimization of representation learning and clustering leads to improved cancer subtype identification.

Conclusions:

• SACR-DCA offers a robust framework for multi-omics cancer subtype clustering.
• The approach advances precision medicine by enabling more accurate cancer subtyping.
• The proposed method provides a significant improvement over current multi-omics clustering techniques.