iProLSTM-FL: A Federated Learning Framework for Promoter Identification using LSTM Networks

Promoter identification is significant for elucidating gene regulation mechanisms, which are central to biological processes such as development and disease progression. Traditional machine learning (ML) and deep learning (DL) approaches have shown considerable potential in promoter prediction. However, they often rely on centralized data, posing significant privacy concerns, or on non-cooperative frameworks that lack generalizability. In this paper, we propose iProLSTM-FL, a federated learning (FL) framework that integrates privacy-preserving techniques with state-of-the-art sequence modeling techniques, leveraging bidirectional long short-term memory networks and attention mechanisms. When evaluated on a benchmark dataset, iProLSTM-FL achieved an accuracy of 92.85%, demonstrating comparable performance to centralized models and existing works while ensuring data privacy and outperforming non-cooperative methods. This work highlights the efficacy of FL in secure genomic analysis, paving the way for scalable, decentralized promoter prediction frameworks.