FedscGen: privacy-preserving federated batch effect correction of single-cell RNA sequencing data

Mohammad Bakhtiari ¹, Stefan Bonn ^2,3,4,5, Fabian Theis ^6,7,8

¹Institute for Computational Systems Biology, University of Hamburg, Hamburg, Germany. mohammad.bakhtiari@uni-hamburg.de.
²Institute of Medical Systems Bioinformatics, Center for Biomedical AI (bAIome), Center for Molecular Neurobiology (ZMNH), University Medical Center Hamburg-Eppendorf, Hamburg, 20246, Germany.
³German Center for Child and Adolescent Health (DZKJ), Partner Site Hamburg, University Medical Center Hamburg-Eppendorf, Hamburg, 20246, Germany.
⁴Hamburg Center for Kidney Health (HCKH), University Medical Center Hamburg-Eppendorf, Hamburg, 20246, Germany.
⁵Hamburg Center for Translational Immunology (HCTI), University Medical Center Hamburg-Eppendorf, Hamburg, 20246, Germany.
⁶School of Life Sciences Weihenstephan, Technical University of Munich, Munich, Germany.
⁷Helmholtz Zentrum München - German Research Center for Environmental Health, Institute of Computational Biology, Neuherberg, Germany.
⁸Department of Mathematics, Technical University of Munich, Munich, Germany.
⁹Institute for Computational Systems Biology, University of Hamburg, Hamburg, Germany.
¹⁰Data Science in Systems Biology, TUM School of Life Sciences, Technical University of Munich, Freising, Germany.
¹¹Department of Mathematics and Computer Science, University of Southern Denmark, Odense, Denmark.

⁰Institute for Computational Systems Biology, University of Hamburg, Hamburg, Germany. mohammad.bakhtiari@uni-hamburg.de.

Genome Biology +

|

July 22, 2025

View abstract on PubMed

Summary

FedscGen offers a privacy-preserving method for correcting batch effects in single-cell RNA sequencing data. This federated approach ensures secure data sharing and collaboration for improved analysis.

Area Of Science

Computational Biology
Genomics
Bioinformatics

Background

Single-cell RNA sequencing (scRNA-seq) data from clinical samples frequently exhibit batch effects, complicating data integration and analysis.
Genomic privacy concerns significantly limit the sharing of valuable clinical scRNA-seq datasets.
Existing methods for batch effect correction may not adequately address privacy requirements for sensitive clinical data.

Purpose Of The Study

To develop a privacy-preserving, communication-efficient federated method for batch effect correction in scRNA-seq data.
To enable secure collaboration and data sharing for scRNA-seq analysis while mitigating batch effects.
To integrate new studies into existing federated learning frameworks for robust batch effect correction.

Main Methods

FedscGen, a federated learning framework built upon the scGen model, incorporating secure multiparty computation for enhanced privacy.
Implementation of federated training and batch effect correction workflows within the FedscGen framework.
Benchmarking FedscGen against existing methods using diverse scRNA-seq datasets, including the Human Pancreas dataset.

Main Results

FedscGen demonstrates competitive performance in batch effect correction, matching the scGen model on key metrics such as NMI, GC, ILF1, ASW_C, kBET, and EBM.
The method successfully supports federated training and the integration of new datasets, proving its versatility.
Validation across diverse datasets confirms the efficacy and robustness of FedscGen.

Conclusions

FedscGen provides a secure and effective solution for addressing batch effects in clinical scRNA-seq data.
The privacy-preserving nature of FedscGen facilitates real-world collaboration and data sharing, overcoming genomic privacy barriers.
As a FeatureCloud app, FedscGen is readily accessible for secure, collaborative scRNA-seq batch effect correction.

Keywords:

Batch effect correction Federated learning Generative models Privacy-preserving computation Secure multiparty computation (SMPC) Single-cell RNA-seq