Unsupervised Cluster Analysis and Gene Marker Extraction of scRNA-seq Data Based On Non-Negative Matrix Factorization

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at single-cell resolution, crucial for understanding cellular processes like mutation and differentiation.
• Identifying cell heterogeneity is a key challenge in scRNA-seq data analysis.

Purpose of the Study:

• To develop a novel unsupervised framework, MscNMF, for robust scRNA-seq data analysis.
• To enhance the identification of cell heterogeneity and subtypes through multi-subspace similarity learning.

Main Methods:

• Proposed a non-negative matrix factorization (NMF) framework named MscNMF.
• MscNMF integrates data decomposition, multi-subspace similarity learning, and similarity fusion.
• The method learns gene and cell features across subspaces, eliminating noise and redundant information.

Main Results:

• MscNMF effectively identifies cell subpopulations and determines the optimal number of clusters and NMF rank.
• Experiments on eight real scRNA-seq datasets demonstrate MscNMF's superior clustering performance.
• The framework successfully extracts biologically relevant genetic markers.

Conclusions:

• MscNMF provides a robust and effective approach for unsupervised scRNA-seq data analysis.
• The multi-subspace similarity learning strategy enhances the accuracy of cell type identification and marker discovery.
• MscNMF offers a valuable tool for researchers studying cell heterogeneity.