Novel Privacy-Preserving Machine-Learning Method Developed for Genomics Data

A research team from King Abdullah University of Science and Technology (KAUST) has developed a machine-learning approach that addresses a significant challenge in medical research. This new method, described in Science Advances, harnesses the power of artificial intelligence to expedite discoveries from genomic data while ensuring the privacy of individuals is protected.

The challenge in medical research lies in utilizing AI, particularly deep learning models, to analyze omics data without compromising private information that could reveal a person's health status. Traditional privacy preservation methods, such as data encryption, introduce significant computational overheads and still risk retaining private information in the trained models. Another method, local training or federated learning, involves breaking the data into smaller packets for model training but can still potentially leak private information.

To address these issues, the KAUST team proposed an innovative approach that integrates a decentralized shuffling algorithm within the differential privacy framework. This method not only enhances model performance but also maintains a high level of privacy protection by resolving trust issues associated with centralized third-party shufflers. The decentralized shuffler achieves a better balance between privacy preservation and model capability, ensuring robust protection against privacy breaches.

The team's privacy-preserving machine-learning approach, dubbed PPML-Omics, was tested on three challenging multi-omics tasks using three representative deep-learning models. The results demonstrated that PPML-Omics not only outperformed other methods in terms of efficiency but also showed resilience against advanced cyberattacks.