Selecting single cell clustering parameter values using subsampling-based robustness metrics.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Single-cell data analysis is crucial for understanding tissue heterogeneity.
• Existing clustering algorithms often require manual parameter tuning, which can be challenging for users.
• Optimizing parameters is vital for identifying biologically relevant cell types from transcriptomic data.

Purpose of the Study:

• To introduce chooseR, a subsampling-based approach for guiding parameter selection in single-cell clustering.
• To provide a method for assessing the robustness and quality of identified cell clusters.
• To simplify the process of analyzing complex single-cell datasets.

Main Methods:

• Implemented a subsampling-based approach called chooseR.
• Utilized bootstrapped iterative clustering across a range of parameters.
• Applied chooseR to two distinct clustering workflows (Seurat and scVI).

Main Results:

• chooseR successfully guided parameter selection for both Seurat and scVI workflows.
• Identified biologically relevant clusters in both human peripheral blood mononuclear cells (PBMC) and mouse spinal cord datasets.
• Generated a 'robustness score' for each cluster, enabling quality assessment.

Conclusions:

• chooseR is a user-friendly tool for parameter selection and cluster robustness assessment in single-cell analysis.
• The tool is flexible and can be applied across various clustering algorithms, workflows, and datasets.
• chooseR enhances the reliability and interpretability of single-cell RNA sequencing data analysis.