Related Experiment Video
Updated: Oct 8, 2025
10:28
Reusable Single Cell for Iterative Epigenomic Analyses
Published on: February 11, 2022
1.4K
Benchmarking atlas-level data integration in single-cell genomics
Malte D Luecken1, M Büttner1, K Chaichoompu1
1Institute of Computational Biology, Helmholtz Zentrum München, German Research Center for Environmental Health, Neuherberg, Germany.
Nature Methods
|December 24, 2021
Summary
Benchmarking 68 data integration methods across 85 batches of single-cell data reveals optimal strategies. Highly variable gene selection enhances performance, while scaling data hinders biological variation preservation.
Area of Science:
- Computational Biology
- Genomics
- Bioinformatics
Background:
- Single-cell atlases aggregate diverse datasets, introducing complex batch effects.
- Effective data integration is crucial for joint analysis of these large-scale datasets.
Purpose of the Study:
- To benchmark various data integration methods and preprocessing strategies for single-cell atlases.
- To identify optimal methods for removing batch effects while preserving biological variation.
Main Methods:
- Evaluated 68 method/preprocessing combinations on 85 batches (>1.2 million cells) of gene expression, ATAC-seq, and simulation data.
- Utilized 14 metrics assessing scalability, usability, batch effect removal, and biological variation conservation.
- Tested on 13 atlas-level integration tasks.
Main Results:
- Highly variable gene selection improved data integration performance.
- Data scaling prioritized batch removal over biological variation.
- scANVI, Scanorama, scVI, and scGen demonstrated strong performance, especially in complex tasks.
- Single-cell ATAC-sequencing integration was sensitive to feature space selection.
Conclusions:
- The study provides a comprehensive benchmark to guide the selection of data integration methods for single-cell atlases.
- A freely available Python module and pipeline can aid in method selection, benchmarking, and development.
