Combining Global-Constrained Concept Factorization and a Regularized Gaussian Graphical Model for Clustering...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates vast amounts of data, revealing cellular heterogeneity.
• Clustering scRNA-seq data is crucial for downstream analysis but is challenging due to high dimensionality, noise, sparsity, and missing data.
• Existing computational methods often have inadequate efficacy and require pre-specified cluster numbers, limiting their real-world applicability.

Purpose of the Study:

• To develop a novel computational method for accurate and robust clustering of scRNA-seq data.
• To address the limitations of existing methods, particularly the need for pre-defined cluster numbers.
• To effectively leverage both global and local information within scRNA-seq datasets.

Main Methods:

• Developed GCFG, a computational method integrating concept factorization for global data properties and regularized Gaussian graphical models for local embedding relationships.
• Designed an iterative optimization algorithm to learn a cell-cell similarity matrix by combining global and local information.
• Utilized the Louvain community discovery algorithm on the learned similarity matrix for single-cell categorization.

Main Results:

• GCFG was evaluated on 14 real-world scRNA-seq datasets.
• Performance was assessed using Accuracy (ACC) and Adjusted Rand Index (ARI).
• GCFG demonstrated superior effectiveness and robustness compared to 17 other competitive clustering methods.

Conclusions:

• GCFG offers an effective and robust approach for clustering scRNA-seq data.
• The method successfully integrates global and local data characteristics for improved categorization.
• GCFG overcomes limitations of existing methods, providing a valuable tool for transcriptomic heterogeneity analysis.