Multi-view contrastive clustering for cancer subtyping using fully and weakly paired multi-omics data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Cancer subtype identification is vital for precision medicine and personalized treatments.
• High-throughput sequencing generates vast multi-omics data, enabling computational cancer subtyping.
• Integrating multi-omics data is challenging due to missing values and weakly paired omics.

Purpose of the Study:

• To develop a novel unsupervised cancer subtyping model.
• To effectively handle weakly paired multi-omics datasets with missing data.
• To improve the accuracy and reliability of cancer subtyping for clinical applications.

Main Methods:

• Proposed Subtype-MVCC, an unsupervised model using graph convolutional networks.
• Employed intra-view and inter-view contrastive learning for feature extraction.
• Utilized a weighted average fusion strategy to unify omics data dimensions.

Main Results:

• Subtype-MVCC outperformed nine leading computational models on benchmark datasets.
• The model demonstrated robust performance in simulations with varying missing data levels.
• Identified subtypes showed clinical relevance and significant survival outcome associations.

Conclusions:

• Subtype-MVCC effectively addresses challenges in multi-omics data integration for cancer subtyping.
• The model offers a reliable and interpretable approach for identifying clinically relevant cancer subtypes.
• This advancement supports the development of more precise cancer treatment and prevention strategies.