CANTAO: guiding clustering and annotation in single-cell RNA sequencing using average overlap

Area of Science:

• Computational Biology
• Genomics
• Immunology

Background:

• Single-cell RNA sequencing (scRNA-seq) uses unsupervised clustering to define cell identities based on transcriptional similarity.
• A single clustering resolution may fail to capture both broad cell populations and smaller, distinct subpopulations simultaneously.
• Robust comparison and annotation of de novo clusters are challenging when cell identities are unknown prior to sequencing.

Purpose of the Study:

• Introduce CANTAO, a novel computational framework for analyzing single-cell RNA sequencing data.
• Develop and validate the average overlap metric for quantifying distances between single-cell clusters.
• Improve the biological interpretation and annotation of de novo clusters in scRNA-seq data.

Main Methods:

• Propose the average overlap metric, which compares ranked lists of differentially expressed genes in a top-weighted manner to define cluster distances.
• Benchmark CANTAO using truth-known datasets with similar yet distinct cell populations.
• Analyze unsorted mouse thymocytes to characterize T-cell development stages.

Main Results:

• CANTAO, using the average overlap metric, consistently and precisely recapitulates true cell identities in benchmark datasets.
• The method successfully identifies minor T-cell populations, such as double-negative (CD4-CD8-) T cells, within unsorted mouse thymocytes.
• CANTAO demonstrates robust and reproducible characterization of single-cell data, clarifying biological interpretation.

Conclusions:

• CANTAO provides a reliable approach for robustly characterizing single-cell data and interpreting cellular identities.
• The average overlap metric enhances the ability to resolve and annotate both major and minor cell populations.
• This framework facilitates a deeper understanding of biological systems, particularly in complex tissues like the thymus.