Network-Based Structural Learning Nonnegative Matrix Factorization Algorithm for Clustering of scRNA-Seq Data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity.
• Current methods often overlook cell-cell interactions, limiting accuracy in cell type and trajectory analysis.

Purpose of the Study:

• To develop a novel network-based algorithm for improved cell type identification and trajectory inference from scRNA-seq data.
• To address the limitations of existing methods by incorporating cell interaction information.

Main Methods:

• Proposed a network-based structural learning nonnegative matrix factorization (SLNMF) algorithm.
• Constructed cell similarity networks and extracted latent features using network topology.
• Imposed structural constraints to preserve network information for enhanced clustering.

Main Results:

• SLNMF demonstrated superior performance across fourteen scRNA-seq datasets.
• Achieved a 15.32% accuracy improvement over fifteen state-of-the-art methods.
• Successfully identified cell trajectories and improved cell type discovery.

Conclusions:

• SLNMF offers an effective strategy for analyzing scRNA-seq data by exploiting cell-cell interactions.
• The network-based approach enhances the accuracy of cell type and trajectory inference.
• The freely available software provides a valuable tool for the research community.