Significance analysis for clustering with single-cell RNA-sequencing data.

Area of Science:

• Computational biology
• Genomics
• Statistical genetics

Background:

• Unsupervised clustering of single-cell RNA-sequencing (scRNA-seq) data is crucial for identifying cell populations.
• Current clustering algorithms often lack statistical rigor, leading to overconfidence in novel cell type discoveries due to unaddressed variability.

Purpose of the Study:

• To develop a statistically rigorous, model-based hypothesis testing approach for scRNA-seq data clustering.
• To provide a method for statistically evaluating the significance of identified cell populations and assessing clusters from any algorithm.
• To extend these methods to account for batch effects in scRNA-seq data.

Main Methods:

• Extended the significance of hierarchical clustering method to incorporate model-based hypothesis testing.
• Developed a framework for statistical assessment of clusters generated by any clustering algorithm.
• Adapted the approach to handle batch structures within scRNA-seq datasets.

Main Results:

• The proposed method demonstrates improved performance compared to popular clustering workflows.
• Identified instances of over-clustering in real-world datasets, such as the Human Lung Cell Atlas.
• Successfully recapitulated experimentally validated cell type definitions in mouse cerebellar cortex data.

Conclusions:

• Statistically evaluating cell populations in scRNA-seq data is essential for accurate cell type discovery.
• The developed approach enhances the reliability of cell clustering by formally addressing statistical uncertainty.
• This method offers a robust tool for analyzing complex scRNA-seq datasets and validating cell type definitions.