Similarity and Dissimilarity Regularized Nonnegative Matrix Factorization for Single-Cell RNA-seq Analysis.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Traditional sequencing methods overlook cellular functions and differentiation roles.
• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at the single-cell level, revealing cellular heterogeneity.
• Unsupervised clustering is a key method for identifying distinct cell subpopulations within scRNA-seq data.

Purpose of the Study:

• To introduce a novel clustering algorithm, Similarity and Dissimilarity Regularized Nonnegative Matrix Factorization (SDCNMF).
• To enhance the analysis of cellular heterogeneity in scRNA-seq data by simultaneously considering cell similarity and dissimilarity.
• To improve the accuracy of subpopulation identification in single-cell studies.

Main Methods:

• Proposing SDCNMF, a nonnegative matrix factorization method incorporating similarity and dissimilarity constraints.
• Applying SDCNMF to low-dimensional representations of scRNA-seq data.
• Evaluating SDCNMF performance against existing clustering methods on multiple scRNA-seq datasets.

Main Results:

• SDCNMF effectively clusters cells by enforcing proximity for similar cells and separation for dissimilar cells in the low-dimensional space.
• The proposed SDCNMF method demonstrated superior performance compared to other clustering techniques across five scRNA-seq datasets.
• Gene markers identified using SDCNMF align with findings from previous biological studies, validating its effectiveness.

Conclusions:

• SDCNMF is a powerful and effective tool for analyzing single-cell RNA sequencing data.
• The method accurately identifies cellular heterogeneity and subpopulations.
• SDCNMF advances the field of single-cell data analysis by providing improved clustering capabilities.