Single cell clustering based on cell-pair differentiability correlation and variance analysis.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell technologies reveal cellular heterogeneity.
• Identifying intercellular transcriptomic differences is crucial in single-cell RNA sequencing (scRNA-seq).

Purpose of the Study:

• To introduce a new cell similarity measure and clustering algorithm for scRNA-seq data analysis.
• To accurately identify cell types and potential biomarkers from noisy scRNA-seq data.

Main Methods:

• Developed a cell similarity measure based on cell-pair differentiability correlation.
• Created a variance analysis-based clustering algorithm named 'Corr' for automatic cluster number determination.
• Compared 'Corr' with state-of-the-art methods using benchmark and real scRNA-seq datasets.

Main Results:

• The 'Corr' algorithm accurately identifies cell types and determines cluster numbers.
• Demonstrated robustness and superiority over existing methods (e.g., SNN-Cliq) using internal and external validation criteria.
• Identified potential cancer biomarkers and confirmed prognostic effectiveness using independent cancer datasets.

Conclusions:

• The proposed 'Corr' algorithm offers an effective approach for analyzing cellular heterogeneity in scRNA-seq data.
• The differentiability vector provides a novel tool for biomarker discovery in cancer research.
• The method shows promise for accurate cell type identification and prognosis analysis.