Decoupled contrastive multi-view clustering with adaptive false negative elimination for cancer subtyping

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Cancer heterogeneity demands precise subtype identification for effective diagnosis and personalized treatment.
• Multi-omics data integration is crucial for revealing distinct molecular characteristics.
• Existing contrastive clustering methods struggle with inter- and intra-view relationships and false negatives.

Purpose of the Study:

• To develop a novel end-to-end self-supervised learning model for improved cancer subtyping.
• To address limitations in capturing multi-omics data relationships and handling false negatives in clustering.
• To enhance personalized cancer therapies through accurate subtyping and biomarker discovery.

Main Methods:

• Proposed Decoupled Contrastive Multi-view Clustering (DCMC) model.
• Employed multi-view clustering with intra- and inter-view contrastive learning.
• Introduced adaptive false negative elimination and pseudo-label rectification.

Main Results:

• DCMC demonstrated superior performance on 10 cancer datasets compared to 19 state-of-the-art methods.
• Identified potential biomarkers through differential expression analysis in Liver Hepatocellular Carcinoma.
• Validated identified cancer subtypes for therapeutic drug responses.

Conclusions:

• DCMC offers an effective approach for cancer subtyping by leveraging multi-omics data.
• The model's ability to handle false negatives and refine representations leads to improved clustering accuracy.
• Findings support the potential of DCMC in advancing personalized cancer medicine.