Date of Publication :7th October 2025

Abstract: Genomic data has very high dimensionality, large volume, and strict privacy needs. These characteristics make centralized machine learning methods unsuitable because of regulatory limits and heavy communication demands. The growing availability of portable DNA sequencers and IoT-enabled biomedical devices allows data collection right at the edge. However, these devices have limited memory, computation, and bandwidth. To tackle this issue, we suggest a federated learning framework for privacy-preserving genomic analysis across distributed nodes like hospitals or edge devices. Each node carries out local preprocessing, reduces dimensionality, and trains lightweight models. They send only parameter updates to a central server. The server combines these updates using Federated Averaging and then distributes the global model back to clients. We conducted experiments on public breast cancer gene expression datasets to assess classification accuracy, communication costs, and runtime under simulated edge conditions. The results indicate that our framework performs similarly to centralized training while reducing communication needs by a large margin and protecting patient privacy. These findings highlight the potential of federated edge learning to support scalable and secure genomic prediction in distributed healthcare systems.

Reference :