On behalf of the amazing teal of ELSA’s Health Privacy Challenge, we want to share some thrilling news:
The Health Privacy Challenge has been successfully completed, and its findings were presented during the Health Privacy Session at #CAMDA, as a part of the ISMB/ECCB Conference on July 24, 2025.
The session began with a keynote by Antti Honkela, who introduced attacks and defense mechanisms, highlighting the recent EDPB opinion on the anonymity of trained machine learning models. He emphasised the importance of designing effective benchmarking strategies and privacy competitions.
Following the keynote, Hakime Öztürk introduced the challenge, a community-driven initiative focused on benchmarking privacy-preserving synthetic data for bulk and single-cell gene expression datasets. She underscored the privacy-utility trade-off based on the baseline methods and received submissions.
A panel of experts, including Antti Honkela, Spiros Denaxas, David P. Kreil, Wenzhong Xiao, and Joaquin Dopazo, addressed critical questions about privacy implementation in healthcare, determining what constitutes “private enough”. They also answered audience questions regarding the applicability of federated learning in the UK, EU, and USA settings, and discussed alignment of benchmarking systems with real-world clinical deployment.
Finally, challenge participants showcased their work. In the bulk RNA-seq track, Steven Golob (University of Washington) evaluated Private-PGM models with different privacy budgets, while Jules Kreuer (University of Tubingen) presented NoisyDiffusion, a privacy-preserving model optimizing both utility and privacy.
In the single-cell RNA-seq track, Andrew Wicks (DKFZ) discussed the application of non-negative matrix factorization (NMF) with differential privacy at various model layers, and Patrick McKeever (University of Washington) provided a comparative analysis of various scRNA-seq generators, including scDesign2 and cfDiffusion, focusing on their trade-offs between utility and privacy.
We thank everyone involved in making the challenge a dynamic learning platform and look forward to future editions!
We introduced the Health Privacy Challenge Session in this article: https://elsa-ai.eu/elsa-health-privacy-challenge/