A novel metric-based meta-learning method to predict synergistic drug combinations

Due to the limitations of monotherapy, combination drug therapy has been widely applied to many refractory diseases that are difficult to treat with a single drug. The use of drugs with different mechanisms of action can enhance therapeutic effects. However, discovering the synergistic effects between drugs is a lengthy process, especially for less commonly used drugs. Previous artificial intelligence models often focused on existing drug combinations and required extensive data support. To explore the potential for new drug combinations from existing ones, we employ meta-learning on knowledge graphs to predict drug combinations. This paper proposes the MetaDCP method, which leverages the relation learner and triple learner to better capture the latent relationships between drugs and diseases, and can effectively handle sparse drug combinations. Compared to other baseline models, MetaDCP consistently outperformed existing methods in terms of multiple metrics. Finally, we validated the practicality of MetaDCP in predicting effective drug combinations for hypertension.