An interpretable framework for clustering single-cell RNA-Seq datasets.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-Seq) generates complex datasets requiring robust analysis.
• Existing unsupervised clustering methods often use non-intuitive hyperparameters and struggle with inherent data subjectivity.
• The definition of "cell type" remains a challenge in scRNA-Seq data interpretation.

Purpose of the Study:

• To introduce DendroSplit, a novel, interpretable framework for scRNA-Seq data analysis.
• To address limitations in clustering interpretability and subjectivity in scRNA-Seq data.
• To enable the discovery of multi-level, biologically meaningful cell populations.

Main Methods:

• DendroSplit utilizes a feature selection approach for clustering scRNA-Seq data.
• The framework is motivated by the biological definition of "cell type" to guide population discovery.
• The method was evaluated on several established scRNA-Seq datasets.

Main Results:

• DendroSplit demonstrates comparable accuracy and speed to existing scRNA-Seq clustering methods.
• The framework successfully uncovers multiple levels of biologically relevant cell populations.
• The analysis highlights the method's efficacy and computational efficiency.

Conclusions:

• DendroSplit offers a significant advancement in interpretable scRNA-Seq data clustering.
• The framework provides a user-friendly and effective alternative for analyzing complex single-cell datasets.
• The DendroSplit software package is publicly available for broader research application.