Minimal gene set discovery in single-cell mRNA-seq datasets with ActiveSVM

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High costs of single-cell messenger RNA sequencing (mRNA-seq) restrict its widespread use in research and clinical settings.
• Targeted sequencing of reduced gene sets offers a potential solution to mitigate these costs by focusing on biologically relevant information.

Purpose of the Study:

• To develop and validate an active learning method for identifying minimal yet highly informative gene sets from single-cell mRNA-seq data.
• To enable accurate cell type, physiological state, and genetic perturbation identification using significantly fewer genes.

Main Methods:

• Implementation of an active learning approach utilizing an active support vector machine (ActiveSVM) classifier for feature selection.
• Application of the ActiveSVM method to diverse single-cell datasets, including cell atlases and disease characterization data.

Main Results:

• The ActiveSVM feature selection successfully identified minimal gene sets capable of achieving approximately 90% accuracy in cell-type classification.
• Demonstrated efficacy across various biological contexts, highlighting the method's robustness.

Conclusions:

• The developed active learning method effectively identifies minimal gene sets for single-cell mRNA-seq analysis.
• This approach has the potential to significantly reduce sequencing costs, thereby facilitating broader applications in clinical diagnostics, drug discovery, and genetic screening.