FedKD-DTI: Drug-Target Interaction Prediction Based on Federated Knowledge Distillation

Drug-target interaction (DTI) prediction is critical in the early stages of drug discovery, narrowing the search space for DTIs and reducing the cost and time required for traditional high-throughput screening. However, DTI-related data are usually distributed across different institutions and their sharing is restricted because of data privacy and intellectual property rights. It is essential to construct a scheme that enhances multi-institutional collaboration to improve prediction accuracy while protecting data privacy. To this end, we propose FedKD-DTI, the first framework based on federated knowledge distillation, to effectively facilitate multi-party DTI collaborative prediction and ensure data privacy and security. FedKD-DTI uses knowledge distillation technology to extract the updated knowledge of all client models, and train an auxiliary model on the server to achieve knowledge aggregation, which can effectively utilize the knowledge contained in public and private data. We evaluate FedKD-DTI on three benchmark datasets and compare it with four baselines. The results show that FedKD-DTI is very close to centralized learning and significantly better than localized learning. Furthermore, FedKD-DTI outperforms federated learning-based baselines on both independent and identically distributed data and non-independent and identically distributed data. Overall, FedKD-DTI improves the DTI prediction while ensuring data security, which promotes institutions collaboration to accelerate the drug discovery process.