Robust classification of single-cell transcriptome data by nonnegative matrix factorization.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptome data offer high resolution for studying cell heterogeneity.
• Unsupervised classification of this data is challenging due to inherent noise.
• Existing methods like Principal Component Analysis (PCA) struggle with noisy single-cell data.

Purpose of the Study:

• To adapt Nonnegative Matrix Factorization (NMF) for identifying subpopulations in single-cell transcriptome data.
• To evaluate NMF's performance against PCA in cell subtyping.
• To develop novel approaches for determining the number of subpopulations.

Main Methods:

• Adapted Nonnegative Matrix Factorization (NMF) using a cell-centered approach to identify 'metacells'.
• Applied NMF to three diverse single-cell transcriptome datasets (RT-qPCR and RNA-seq).
• Compared NMF performance against Principal Component Analysis (PCA).

Main Results:

• NMF demonstrated superior accuracy and robustness in identifying subpopulations compared to PCA.
• NMF successfully identified broad cell classes in large-scale datasets.
• The method enables direct and unbiased identification of feature genes.
• Novel approaches were proposed for determining the number of subpopulations.

Conclusions:

• Nonnegative Matrix Factorization (NMF) is a robust, informative, and simple method for unsupervised cell subtyping.
• NMF offers an effective alternative to PCA for analyzing single-cell gene expression data.
• The cell-centered NMF approach provides valuable insights into cellular heterogeneity.