Consensus prediction of cell type labels in single-cell data with popV

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell sequencing analysis requires accurate cell-type classification.
• Existing cell-type label transfer methods lack robust uncertainty estimation.
• Uncertainty quantification is vital for interpretable and reliable single-cell data analysis.

Purpose of the Study:

• To introduce popular Vote (popV), an ensemble-based method for cell-type classification.
• To provide accurate cell-type labels with associated uncertainty scores.
• To streamline the manual inspection process in single-cell data annotation.

Main Methods:

• Developed popV, an ensemble of prediction models utilizing an ontology-based voting scheme.
• Applied popV to annotated reference atlases for transferring labels to unannotated query datasets.
• Evaluated popV's performance and uncertainty estimation capabilities through multiple case studies.

Main Results:

• PopV achieves accurate cell-type labeling in single-cell sequencing data.
• The method provides reliable uncertainty scores for cell-type annotations.
• PopV confidently annotates most cells while identifying challenging populations for manual review.

Conclusions:

• PopV enhances the interpretability and utility of cell-type annotations in single-cell analysis.
• The uncertainty scores generated by popV aid in focusing manual inspection efforts.
• PopV streamlines the overall annotation process by reducing the burden of manual data curation.