A benchmark of batch-effect correction methods for single-cell RNA sequencing data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data present batch effects from different technologies, challenging data integration.
• Effective batch correction is crucial for analyzing the growing volume of scRNA-seq data.

Purpose of the Study:

• To benchmark and compare available computational methods for batch-effect removal in scRNA-seq data.
• To identify the most suitable methods for accurate and efficient data integration.

Main Methods:

• Evaluated 14 batch correction algorithms on scRNA-seq datasets.
• Assessed methods based on computational runtime, scalability, and batch-effect removal efficacy.
• Utilized metrics like kBET, LISI, ASW, and ARI across five distinct experimental scenarios.
• Investigated the impact of batch correction on differential gene expression analysis.

Main Results:

• Compared 14 batch correction methods across various scenarios including identical/non-identical cell types, multiple batches, large datasets, and simulated data.
• Performance was evaluated using kBET, LISI, ASW, and ARI metrics.
• Harmony, LIGER, and Seurat 3 demonstrated strong performance in batch integration and cell type purity preservation.

Conclusions:

• Harmony, LIGER, and Seurat 3 are recommended for scRNA-seq data integration.
• Harmony is the preferred initial choice due to its superior computational efficiency.
• The study provides guidance for selecting appropriate batch correction tools for large-scale transcriptomic data analysis.