COMSE: analysis of single-cell RNA-seq data using community detection-based feature selection.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers individual cell analysis but faces challenges with high gene dimensions and low cell numbers.
• Only a subset of detected genes are relevant for cell-type specific functions, necessitating effective feature selection.

Purpose of the Study:

• To present COMSE, an unsupervised feature selection framework for scRNA-seq data.
• To improve the identification of homogenous cell substates and enhance cell clustering accuracy.
• To enable robust analysis of scRNA-seq datasets from diverse sources, including batch effect correction.

Main Methods:

• COMSE utilizes community detection algorithms for unsupervised feature selection from scRNA-seq data.
• The framework identifies informative gene communities associated with biological and technical variations.
• Evaluations involved real and simulated scRNA-seq datasets to assess performance in cell clustering and batch effect handling.

Main Results:

• COMSE successfully identified homogenous cell substates with high resolution, distinguishing cell cycle stages.
• The framework demonstrated superior performance in cell clustering compared to existing methods, even with high dropout rates.
• COMSE effectively parsed biological signals from technical noise (batch effects), facilitating integrated analysis of multi-source scRNA-seq data.

Conclusions:

• COMSE is an efficient unsupervised framework for selecting informative genes in scRNA-seq, enhancing cell sub-state identification and clustering.
• The identified gene subsets reveal biological and technical heterogeneity, supporting applications like batch effect correction and pathway analysis.
• COMSE also demonstrates robust performance for bulk RNA-seq data analysis.