Community benchmarking of generative models for privacy-preserving synthetic bulk RNA-seq data generation
Abstract
Background: The synthesis of anonymized data derived from real-world cohorts offers a promising strategy for regulatory-compliant and privacy-preserving biological data sharing, potentially facilitating model development that can improve predictive performance. However, the extent to which generative models can preserve biological signals while remaining resilient to adversarial privacy attacks in high-dimensional omics contexts remains underexplored. To address this gap, the CAMDA 2025 Health Privacy Challenge launched a community-driven effort to evaluate synthetic and privacy-preserving data generation for bulk RNA-seq cohorts.
Results: Building on this initiative, we systematically benchmark 11 generative methods across two cancer cohorts (~ 1,000 and ~ 5,000 patients) over 978 landmark genes. Methods are evaluated across distributional fidelity, downstream utility, biological plausibility and empirical privacy risk, with emphasis on trade-offs between vulnerability to membership inference attacks and other evaluation dimensions. Expressive deep generative models achieve strong predictive utility and differential expression recovery, but are often more vulnerable to membership inference risk under the evaluated threat model. Differentially private methods improve resistance to attacks at the cost of reduced utility, while simpler statistical approaches offer competitive utility with moderate privacy risk and fast training.
Conclusions: Synthetic bulk RNA-seq quality is inherently multi-dimensional and shaped by trade-offs between utility, biological preservation and privacy. Our results indicate that model architecture drives distinct trade-offs across these axes, suggesting that model choice should align with dataset characteristics, intended downstream use and privacy requirements. Privacy risk should also be assessed using multiple complementary attack methods and, where possible, mitigated through formal differential privacy protection.
Related
- RNAGEN: A Generative Adversarial Network‐Based Model to Generate Synthetic RNA Sequences to Target Proteins
- Structure-Preserving Excess Gibbs Learning for Multicomponent Phase Equilibria: A Falsifiable and Uncertainty-Aware Synthetic Study
- PathEdit: Diagnostic-Preserving Counterfactual Medical Image Editing through State-Factorized Generative Intervention
- A data-driven modeling framework for mapping genotypes to synthetic microbial community functions
- Benchmarking cell segmentation in two Spatialomics platforms with open-source software
BibTeX
@article{ozturk2026community,
title = {{Community benchmarking of generative models for privacy-preserving synthetic bulk RNA-seq data generation}},
author = {Hakime Öztürk and Tejumade Afonja and Joonas Jälkö and Ruta Binkyte and Pablo Rodriguez-Mier and Sebastian Lobentanzer and Andrew Wicks and Jules Kreuer and Sofiane Ouaari and Nico Pfeifer and Shane Menzies and Sikha Pentyala and Daniil Filienko and Steven Golob and Patrick McKeever and Jineta Banerjee and Luca Foschini and Martine De Cock and Julio Saez-Rodriguez and Mario Fritz and Oliver Stegle and Antti Honkela},
journal = {Genome Biology},
year = {2026},
month = nov,
doi = {10.1186/s13059-026-04326-1},
url = {https://link.springer.com/article/10.1186/s13059-026-04326-1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Community benchmarking of generative models for privacy-preserving synthetic bulk RNA-seq data generation
AU - Hakime Öztürk
AU - Tejumade Afonja
AU - Joonas Jälkö
AU - Ruta Binkyte
AU - Pablo Rodriguez-Mier
AU - Sebastian Lobentanzer
AU - Andrew Wicks
AU - Jules Kreuer
AU - Sofiane Ouaari
AU - Nico Pfeifer
AU - Shane Menzies
AU - Sikha Pentyala
AU - Daniil Filienko
AU - Steven Golob
AU - Patrick McKeever
AU - Jineta Banerjee
AU - Luca Foschini
AU - Martine De Cock
AU - Julio Saez-Rodriguez
AU - Mario Fritz
AU - Oliver Stegle
AU - Antti Honkela
PY - 2026
DA - 2026/11/09
JO - Genome Biology
DO - 10.1186/s13059-026-04326-1
UR - https://link.springer.com/article/10.1186/s13059-026-04326-1
AB - Background: The synthesis of anonymized data derived from real-world cohorts offers a promising strategy for regulatory-compliant and privacy-preserving biological data sharing, potentially facilitating model development that can improve predictive performance. However, the extent to which generative models can preserve biological signals while remaining resilient to adversarial privacy attacks in high-dimensional omics contexts remains underexplored. To address this gap, the CAMDA 2025 Health Privacy Challenge launched a community-driven effort to evaluate synthetic and privacy-preserving data generation for bulk RNA-seq cohorts. Results: Building on this initiative, we systematically benchmark 11 generative methods across two cancer cohorts (~ 1,000 and ~ 5,000 patients) over 978 landmark genes. Methods are evaluated across distributional fidelity, downstream utility, biological plausibility and empirical privacy risk, with emphasis on trade-offs between vulnerability to membership inference attacks and other evaluation dimensions. Expressive deep generative models achieve strong predictive utility and differential expression recovery, but are often more vulnerable to membership inference risk under the evaluated threat model. Differentially private methods improve resistance to attacks at the cost of reduced utility, while simpler statistical approaches offer competitive utility with moderate privacy risk and fast training. Conclusions: Synthetic bulk RNA-seq quality is inherently multi-dimensional and shaped by trade-offs between utility, biological preservation and privacy. Our results indicate that model architecture drives distinct trade-offs across these axes, suggesting that model choice should align with dataset characteristics, intended downstream use and privacy requirements. Privacy risk should also be assessed using multiple complementary attack methods and, where possible, mitigated through formal differential privacy protection.
SN - 1474-760X
ER -