EC-PGMGR: Ensemble Clustering Based on Probability Graphical Model With Graph Regularization for Single-Cell RNA-seq...

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Single-cell RNA sequencing (scRNA-seq) generates large datasets requiring robust clustering for cellular phenotype identification.
• Existing individual clustering methods have limitations in capturing data aspects and depend on parameter settings.
• Ensemble clustering methods offer improved performance but can be dependent on base clustering results and lack self-regulation.

Purpose of the Study:

• To develop a novel, robust unsupervised ensemble clustering algorithm for scRNA-seq data.
• To address limitations of individual and existing ensemble clustering methods.
• To enhance the accuracy and reliability of cellular phenotype identification from scRNA-seq data.

Main Methods:

• Proposed a novel Ensemble Clustering algorithm based on Probability Graphical Model with Graph Regularization (EC-PGMGR).
• Utilized parameter controlling in Probability Graphical Model (PGM) for automatic cluster number determination.
• Incorporated a regularization term and a pre-learning process for self-regulation of base clustering results.

Main Results:

• EC-PGMGR demonstrated superior performance compared to 4 individual and 2 ensemble clustering methods across 7 datasets.
• Evaluated using Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI), showing improvements in accuracy, robustness, and effectiveness.
• The algorithm successfully integrated diverse clustering results, enhancing active methods and weakening inactive ones.

Conclusions:

• EC-PGMGR provides an effective and robust approach for integrating clustering results in scRNA-seq data analysis.
• The method enhances accuracy and reliability for downstream biological analyses and cellular phenotype identification.
• Offers potential new insights for clustering applications in other scientific domains.