SMURF: embedding single-cell RNA-seq data with matrix factorization preserving self-consistency.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cell-specific transcriptomic analysis for developmental biology, disease, and cancer research.
• scRNA-seq data is characterized by sparsity due to 'dropout' events, necessitating computational solutions like imputation.

Purpose of the Study:

• To introduce SMURF, a novel tool for extracting low-dimensional cell and gene embeddings from scRNA-seq data.
• To evaluate SMURF's performance in cell subpopulation discovery, cell cycle analysis, and gene expression recovery.

Main Methods:

• Matrix factorization with a mixture of Poisson-Gamma divergence as the objective function.
• Preservation of self-consistency in extracted embeddings.
• Application to in silico and eight wet-lab scRNA datasets.

Main Results:

• SMURF demonstrated effective cell subpopulation discovery on multiple datasets.
• The tool successfully reduced cell embeddings to a 1D-oval space to recover cell cycle dynamics.
• SMURF exhibited robust gene expression recovery, surpassing imputation methods in silico, with low RMSE and high Pearson correlation.

Conclusions:

• SMURF provides a powerful alternative to imputation for analyzing scRNA-seq data, offering robust cell embedding and gene expression recovery.
• The tool's ability to identify cell subpopulations and reconstruct biological processes like cell cycle makes it valuable for transcriptomic studies.