Detecting heterogeneity in single-cell RNA-Seq data by non-negative matrix factorization.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA-Sequencing (scRNA-Seq) provides high-resolution biological insights but lacks adequate bioinformatics tools for data analysis.
• Analyzing heterogeneous scRNA-Seq data is crucial for understanding complex biological systems.

Purpose of the Study:

• To evaluate the performance of Non-negative Matrix Factorization (NMF) for analyzing diverse scRNA-Seq datasets.
• To compare NMF with other unsupervised clustering methods like K-means and hierarchical clustering.
• To identify novel gene markers and biological modules within heterogeneous cell populations.

Main Methods:

• Application of Non-negative Matrix Factorization (NMF) to various scRNA-Seq datasets, including mouse hematopoietic stem cells and human glioblastoma.
• Comparative analysis of NMF against K-means and hierarchical clustering for accuracy in separating cell groups.
• Gene ranking using importance scores (D-scores) to identify key markers.
• Integration of NMF with the FEM modularity detection method to uncover protein-protein interaction modules.

Main Results:

• NMF demonstrated higher accuracy in separating distinct cell groups across multiple scRNA-Seq datasets compared to K-means and hierarchical clustering.
• NMF uniquely identified genes expressed at intermediate levels as top-ranked markers for cell subpopulations.
• The combination of NMF and FEM successfully revealed biologically relevant protein-protein interaction modules.

Conclusions:

• Non-negative Matrix Factorization (NMF) is a robust and accurate method for analyzing heterogeneous single-cell RNA-Sequencing data.
• NMF facilitates the discovery of cell subpopulations and key gene markers, enhancing our understanding of biological complexity.
• The NMF-based package offers a valuable tool for scRNA-Seq data analysis and module discovery.