Multi-View Clustering With Graph Learning for scRNA-Seq Data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution gene expression data.
• Analyzing scRNA-seq data for cell type distribution faces challenges due to heterogeneity, high dimensionality, and noise.
• Existing methods struggle to comprehensively capture cell characteristics from a single feature space.

Purpose of the Study:

• To develop a novel algorithm for improved cell type distribution analysis in scRNA-seq data.
• To address the limitations of existing methods in handling scRNA-seq data complexity.
• To enhance the accuracy and robustness of cell type clustering.

Main Methods:

• Proposed a multi-view clustering with graph learning (MCGL) algorithm.
• Integrated multi-view learning to construct multiple feature spaces for comprehensive cell characterization.
• Employed adaptive graph learning to capture dynamic cell-cell similarities and topological relationships.
• Developed a unified optimization framework for simultaneous graph learning, factorization, and cell-type clustering.

Main Results:

• MCGL effectively characterizes scRNA-seq data from multiple perspectives.
• The algorithm adaptively learns cell similarity graphs, overcoming limitations of fixed similarity assumptions.
• MCGL decomposes cell networks into view-specific and common components, improving topological relationship characterization.
• Validated on ten diverse scRNA-seq datasets, MCGL significantly outperformed fourteen state-of-the-art algorithms.

Conclusions:

• MCGL provides a robust and effective approach for scRNA-seq data analysis.
• The multi-view and adaptive graph learning strategies enhance cell type clustering accuracy.
• MCGL represents a significant advancement in analyzing complex single-cell gene expression data.