Multi-view manifold regularized compact low-rank representation for cancer samples clustering on multi-omics data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer type identification is crucial for early diagnosis and treatment.
• Clustering cancer samples based on expression patterns aids in identifying distinct cancer types.
• Integrating multi-omics data offers advantages over single-omics approaches for cancer clustering but faces challenges due to data heterogeneity and noise.

Purpose of the Study:

• To develop an advanced method for cancer clustering by extracting complementary information from multi-omics data.
• To improve the accuracy and reliability of cancer type identification through integrated data analysis.

Main Methods:

• Proposed a novel low-rank subspace clustering method: multi-view manifold regularized compact low-rank representation (MmCLRR).
• Treated each omics data as a distinct view and enforced consistency constraints on low-rank affinity matrices.
• Incorporated manifold regularization and concept factorization for enhanced subspace learning and dictionary updating.
• Utilized a linearized alternating direction method with adaptive penalty for optimization.

Main Results:

• MmCLRR successfully extracts complementary information from multi-omics data for cancer clustering.
• The method learns a consistent subspace representation across different omics data views.
• Concept factorization enables dynamic dictionary updates, boosting subspace learning capabilities.

Conclusions:

• MmCLRR demonstrates superior performance in cancer sample clustering compared to existing multi-view methods.
• The proposed method effectively addresses the challenges of high heterogeneity and noise in multi-omics data.
• This approach enhances the identification of cancer types through integrated analysis.