Single-cell RNA-seq data normalization: A benchmarking study.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Technical factors introduce noise and biases in single-cell RNA sequencing (scRNA-seq) data.
• Normalization is crucial for accurate scRNA-seq data analysis.
• Selecting appropriate normalization methods is challenging due to various available tools.

Purpose of the Study:

• To benchmark and compare the performance of six widely used normalization methods for scRNA-seq data.
• To evaluate normalization methods based on cell clustering, differential expression analysis, and computational resource usage.
• To provide guidance for researchers in choosing the most suitable normalization tool for their specific scRNA-seq datasets.

Main Methods:

• Benchmarking analysis of six normalization methods.
• Evaluation using seven real and four simulated scRNA-seq datasets.
• Assessment criteria included cell clustering accuracy, differential expression analysis performance, and computational resource requirements.

Main Results:

• Dino demonstrated superior performance for clustering large datasets (10x Genomics) and datasets with many cells.
• scTransform showed strong performance for datasets generated using full-length library preparation protocols.
• SCnorm was identified as a suitable method for small-scale scRNA-seq datasets.

Conclusions:

• The choice of normalization method significantly impacts scRNA-seq data analysis outcomes.
• Specific methods like Dino, scTransform, and SCnorm are recommended for particular dataset characteristics.
• This study provides a valuable reference for selecting normalization tools to improve the accuracy and reliability of scRNA-seq analyses.