Evaluation of Cell Type Annotation R Packages on Single-cell RNA-seq Data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Automated cell type annotation is crucial for single-cell RNA sequencing (scRNA-seq) data analysis.
• Existing supervised and semi-supervised methods require comprehensive evaluation for scRNA-seq applicability.
• Adaptability of bulk omics data analysis methods to scRNA-seq remains unclear.

Purpose of the Study:

• To systematically evaluate and compare ten publicly available cell type annotation methods for scRNA-seq data.
• To assess method performance across diverse datasets, including simulation data, focusing on accuracy, robustness, and rare cell type detection.
• To determine the suitability of methods developed for other omics data, like DNA methylation, for scRNA-seq analysis.

Main Methods:

• Evaluated ten R packages: eight scRNA-seq specific (Seurat, scmap, SingleR, CHETAH, SingleCellNet, scID, Garnett, SCINA) and two repurposed (CP, RPC).
• Conducted systematic comparisons on multiple public scRNA-seq and simulation datasets.
• Assessed accuracy (intra/inter-dataset), robustness (gene filtering, cell type similarity, class number), and rare/unknown cell type detection.

Main Results:

• Seurat, SingleR, CP, RPC, and SingleCellNet demonstrated strong overall performance.
• Seurat excelled in annotating major cell types and showed robustness against downsampling.
• Seurat struggled with rare cell populations and highly similar cell types compared to SingleR and RPC.

Conclusions:

• Several methods, including Seurat, SingleR, CP, RPC, and SingleCellNet, are effective for scRNA-seq cell type annotation.
• Seurat is optimal for major cell types but less effective for rare or similar cell types.
• CP and RPC show promise as adaptable methods from other omics data for scRNA-seq analysis.