Analyzing Single Cell RNA Sequencing with Topological Nonnegative Matrix Factorization.

Area of Science:

• Computational Biology
• Data Science
• Statistics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional, complex data, attracting significant interest in computational biology and data science.
• Nonnegative Matrix Factorization (NMF) is a dimensionality reduction technique offering meta-gene interpretation but lacks multiscale analysis capabilities.

Purpose of the Study:

• Introduce novel Nonnegative Matrix Factorization (NMF) methods for enhanced single-cell RNA sequencing (scRNA-seq) data analysis.
• Address the limitations of existing NMF approaches, specifically the lack of multiscale analysis.

Main Methods:

• Developed two persistent Laplacian regularized NMF methods: topological NMF (TNMF) and robust topological NMF (rTNMF).
• Evaluated TNMF and rTNMF on 12 diverse scRNA-seq datasets.
• Utilized TNMF and rTNMF for visualization with Uniform Manifold Approximation and Projection (UMAP) and t-distributed Stochastic Neighbor Embedding (t-SNE).

Main Results:

• TNMF and rTNMF significantly outperformed existing NMF-based methods across all tested datasets.
• The proposed methods provide effective multiscale analysis for scRNA-seq data.
• TNMF and rTNMF demonstrated utility in enhancing data visualization techniques like UMAP and t-SNE.

Conclusions:

• TNMF and rTNMF represent significant advancements in NMF for scRNA-seq data analysis.
• These methods offer improved performance and novel capabilities for exploring complex biological data.
• The topological regularization enhances the interpretability and robustness of NMF components for scRNA-seq.