Clustering trees: a visualization for evaluating clusterings at multiple resolutions.

Area of Science:

• Computational Biology
• Data Science
• Bioinformatics

Background:

• Clustering is essential for analyzing large datasets, particularly in single-cell RNA sequencing (scRNA-seq) for cell type identification.
• The number of clusters and algorithm results can vary significantly based on chosen parameters, complicating data interpretation.
• Determining the optimal clustering resolution is a common challenge in scRNA-seq data analysis.

Purpose of the Study:

• To introduce 'clustering trees' as a novel visualization method to explore the impact of varying clustering resolutions.
• To provide a tool that allows researchers to observe sample migration across clusters as resolution changes.
• To facilitate informed selection of clustering resolution by enabling overlay of meta-information.

Main Methods:

• Development of the 'clustree' R package for generating clustering trees.
• Application of clustering trees to simulated datasets.
• Validation using the classical iris dataset and a complex scRNA-seq dataset.

Main Results:

• Clustering trees effectively illustrate the hierarchical relationships between clusters at multiple resolutions.
• The visualization aids in understanding how individual samples or cells transition between clusters as resolution increases.
• Meta-information overlay successfully informed resolution choice and cluster identification in real-world datasets.

Conclusions:

• Clustering trees offer a powerful approach to navigate and interpret clustering results across different resolutions.
• The 'clustree' R package provides an accessible tool for researchers analyzing large-scale biological data, especially scRNA-seq.
• This visualization method enhances the interpretability of clustering analyses and aids in robust cell type discovery.