A comparison of automatic cell identification methods for single-cell RNA sequencing data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptomics enables detailed cellular composition analysis.
• Manual cell identification is a bottleneck, being time-consuming and irreproducible.
• Automated cell identification methods are needed due to increasing data scale.

Purpose of the Study:

• To benchmark various automated cell identification methods for single-cell RNA sequencing data.
• To evaluate classifier performance across diverse datasets and experimental conditions.
• To identify the most effective methods for reproducible cell identity assignment.

Main Methods:

• Benchmarking 22 automated cell identification classifiers (single-cell specific and general-purpose).
• Performance evaluation on 27 diverse single-cell RNA sequencing datasets.
• Assessment using intra-dataset and inter-dataset predictions, accuracy, unclassified cell rates, and computation time.

Main Results:

• Most classifiers performed well, with accuracy decreasing for complex datasets and deep annotations.
• General-purpose support vector machine classifiers demonstrated superior overall performance.
• Sensitivity analyses revealed performance variations based on input features and cell population size.

Conclusions:

• A comprehensive evaluation of automated cell identification methods for single-cell RNA sequencing is presented.
• Open-source code and a Snakemake workflow are provided for reproducible benchmarking and method extension.
• The study facilitates the adoption of robust automated methods for cell identity assignment.