jSRC: a flexible and accurate joint learning algorithm for clustering of single-cell RNA-sequencing data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables the study of cellular heterogeneity and dynamics by profiling gene expression at the individual cell level.
• Large-scale scRNA-seq necessitates efficient clustering algorithms for cell type identification and gene discovery.
• Existing scRNA-seq clustering methods often lack desired accuracy, scalability, and interpretability.

Purpose of the Study:

• To address limitations in current scRNA-seq clustering algorithms.
• To develop an integrated approach for dimension reduction and clustering.
• To enhance the accuracy, scalability, and interpretability of scRNA-seq data analysis.

Main Methods:

• Developed a joint learning algorithm, jSRC (joint sparse representation and clustering), integrating dimension reduction (DR) and clustering.
• Employed DR for scalability and joint learning for improved accuracy.
• Utilized sparse representation on features, assuming similar expression patterns within cell types, to enhance interpretability.

Main Results:

• jSRC significantly outperformed 12 state-of-the-art methods on 15 diverse scRNA-seq datasets, showing an average improvement of 20.29% in various measurements.
• The algorithm demonstrated reduced running time compared to existing methods.
• jSRC accurately identified dynamic cell types associated with COVID-19 progression and proved robust across different datasets.

Conclusions:

• jSRC offers an effective and efficient strategy for analyzing scRNA-seq data, improving cell type identification and interpretability.
• The integrated approach of DR and sparse representation-based clustering provides superior performance and scalability.
• The developed jSRC method is a valuable tool for biological research, including the study of disease progression.