Rarity: discovering rare cell populations from single-cell imaging data

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Cell type identification is crucial for single-cell data analysis.
• Unsupervised clustering aids in discovering known and unknown cell populations.
• Rare cell types pose challenges due to weak expression signatures.

Purpose of the Study:

• To develop a robust framework for identifying rare cell types in single-cell data.
• To improve the consistency and interpretability of unsupervised clustering for rare cell discovery.
• To address the limitations of typical unsupervised methods in detecting rare subpopulations.

Main Methods:

• Developed a novel statistical framework named Rarity for unsupervised clustering.
• Employed a Bayesian latent variable model for cell clustering.
• Assigned cells to inferred latent binary on/off expression profiles.

Main Results:

• Rarity enables more robust and consistent discovery of rare cell types.
• The framework offers increased sensitivity to rare cell populations.
• Rarity allows for control and interpretation of potential false positive discoveries.

Conclusions:

• Rarity provides a powerful tool for identifying rare cell types in single-cell datasets.
• The Bayesian approach enhances the detection and understanding of cellular heterogeneity.
• The method is demonstrated on various imaging mass cytometry (IMC) datasets.