GUIDING CLUSTERING AND ANNOTATION IN SINGLE-CELL RNA SEQUENCING USING THE AVERAGE OVERLAP METRIC

Area of Science:

• Computational Biology
• Genomics
• Immunology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cell type definition via unsupervised clustering.
• A single clustering resolution struggles to capture both broad populations and rare subpopulations simultaneously.
• Annotating de novo clusters is challenging when cell identities are unknown prior to sequencing.

Purpose of the Study:

• To develop a robust method for comparing and annotating de novo cell clusters derived from scRNA-seq data.
• To define a distance metric between single-cell clusters for accurate biological interpretation.
• To address the challenge of identifying both major and minor cell populations in complex datasets.

Main Methods:

• Proposed the average overlap metric to compare ranked lists of differentially expressed genes between clusters.
• Benchmarked the approach on a known dataset of distinct T-cell populations.
• Applied the method to unsorted mouse thymocyte data to characterize T-cell development stages.

Main Results:

• The average overlap metric demonstrated consistent, precise, and biologically meaningful recapitulation of true cell identities in a known dataset.
• Successfully characterized T-cell development stages in mouse thymus, including difficult-to-detect double-negative (CD4-CD8-) T-cells.
• Showcased the ability of average overlap to enable robust and reproducible characterization of single cells in highly homogeneous populations.

Conclusions:

• Measuring cluster similarity using average overlap of marker gene rankings provides a robust method for single-cell data analysis.
• This approach enhances the biological interpretation of cell identities, particularly within complex and homogeneous cell populations.
• The method facilitates the confident detection and characterization of minor cell populations in scRNA-seq data.