Revealing the Best Strategies for Rare Cell Type Detection in Multi-Sample Single-Cell Datasets.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity and rare cell types.
• Existing rare cell detection methods often struggle in multi-sample settings due to batch effects and data imbalance.
• Understanding performance in multi-sample scRNA-seq is crucial for rare cell discovery.

Purpose of the Study:

• To systematically evaluate rare cell detection methods in multi-sample scRNA-seq data.
• To compare different analytical strategies for rare cell detection.
• To identify optimal methods and workflows for large-scale studies.

Main Methods:

• Benchmarking of five rare cell detection tools (CellSIUS, GapClust, GiniClust, scCAD, SCISSORS) and a scGPT-based method.
• Evaluation under three strategies: individual sample, pooled sample, and batch-corrected pooled sample detection.
• Performance assessment across multiple public scRNA-seq datasets using standardized metrics.

Main Results:

• Batch-corrected pooled sample detection consistently yielded the highest performance.
• Individual sample detection strategies showed the weakest results.
• scCAD demonstrated robust and stable performance across various datasets and conditions.

Conclusions:

• Batch correction and pooled analysis are vital for enhancing rare cell detection accuracy.
• This study provides a strategy-level comparison for multi-sample rare cell detection.
• Offers practical guidance for selecting methods and workflows in large-scale scRNA-seq studies.