EC-PGMGR

EC-PGMGR integrates multiple clustering methods with a probability graphical model and graph regularization to identify cell populations in single-cell RNA-seq (scRNA-seq) data.

Key Features:

• Ensemble Clustering Approach: Combines more than two base clustering methods to produce integrated clustering results that capture complementary aspects of scRNA-seq data.
• Probability Graphical Model (PGM) Integration: Employs a PGM to automatically determine the optimal number of clusters without requiring prior knowledge.
• Graph Regularization: Incorporates a regularization term to mitigate the influence of weak base clustering results by assigning weights through a pre-learning process.
• Self-Regulation Mechanism: Uses a self-regulating weighting scheme that enhances active clustering methods while diminishing the impact of less effective ones.
• Performance Evaluation: Validated against four individual clustering methods and two other ensemble methods using Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI).

Scientific Applications:

• Cell type identification: Delineates cellular phenotypes and explores cellular heterogeneity from scRNA-seq datasets.
• Robust clustering where cluster number is unknown: Provides ensemble-based clustering solutions for analyses that require automatic determination of cluster number and integration of multiple methods.

Methodology:

Assembles integrative results from base clustering methods in combination form with self-regulation weights via a pre-learning process; validated on seven datasets generated by various platforms containing between 822 and 5,132 single cells and compared to four individual clustering methods and two ensemble methods using ARI and NMI.