Multi-View Random-Walk Graph Regularization Low-Rank Representation for Cancer Clustering and Differentially...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cancer genome data is complex, comprising multiple views from diverse sources.
• Existing low-rank representation (LRR) methods analyze multi-omics data but struggle with local data geometry.
• Multi-view learning captures view consistency and complementarity but can miss intricate data structures.

Purpose of the Study:

• To propose a novel method, Multi-view Random-walk Graph regularization Low-Rank Representation (MRGLRR), for comprehensive multi-view cancer genomics data analysis.
• To enhance the mining of local geometric information and topological data structures.
• To develop an effective feature gene selection strategy for multi-view cancer data.

Main Methods:

• Utilizing a multi-view model to identify common centroids across data views.
• Constructing a joint affinity matrix for low-rank subspace representation of multi-set data.
• Incorporating random walk graph regularization on KNN graphs to derive accurate sample similarity and retain local geometric information.

Main Results:

• The MRGLRR method effectively extracts hidden information from each view.
• Random walk graph regularization preserves local geometric features and improves topological structure learning.
• The proposed feature gene selection strategy identifies valuable differentially expressed genes.

Conclusions:

• MRGLRR outperforms existing methods in clustering cancer multi-omics data.
• The method demonstrates superior performance in feature gene selection for cancer research.
• MRGLRR offers a more robust approach to analyzing complex, multi-view cancer genomics datasets.