AnnQ: reference-based quantification of cellular abnormality at single-cell resolution.

Area of Science:

• Single-cell RNA sequencing analysis
• Computational biology
• Systems biology

Background:

• Reference-based annotation is standard for cell type assignment in single-cell RNA sequencing (scRNA-seq).
• Ambiguous cell annotations (low confidence, high entropy) are often discarded as artifacts.
• These uncertain cells may represent biologically significant deviations, especially in perturbation studies.

Purpose of the Study:

• To introduce AnnQ (Annotation Quantification of cellular identity uncertainty), a Python framework.
• To repurpose annotation uncertainty as a quantitative measure of cellular abnormality.
• To detect aberrant cellular states missed by conventional analyses.

Main Methods:

• AnnQ extracts uncertainty-aware features from probabilistic cell type assignments.
• Features include confidence, confidence gap, admixture ratio, and entropy.
• An out-of-reference (OOR) score quantifies deviation from reference populations in uncertainty space.

Main Results:

• AnnQ successfully identified aberrant cellular states in genetic perturbation and drug resistance datasets.
• OOR scores detected biological deviations not resolved by standard clustering or differential abundance analyses.
• The framework provides a novel approach for characterizing transitional and abnormal cell states.

Conclusions:

• AnnQ effectively leverages annotation uncertainty to identify biologically relevant cell states.
• The framework complements existing methods for single-cell data analysis.
• AnnQ offers a valuable tool for researchers studying cellular heterogeneity and responses to perturbations.