MuRL-DTI: A Multimodal Feature Fusion Reinforcement Learning Approach for Cold Start in Drug-Target Interactions

Abstract

Drug Target Interaction (DTI) focuses on exploring the interactions between specific drug molecules and their biological targets to assess the efficacy and safety of drugs. Significant advancements have been made in integrating computational techniques compared to traditional approaches, including machine learning-based DTI models, deep learning-based DTI models, and knowledge graph-based DTI models. However, these models often depend on rich contextual and structural information, which can lead to reduced prediction accuracy when dealing with sparse data and pose challenges in cold start scenarios. To address this limitation, we propose a novel DTI prediction method called MuRL-DTI. Specifically, we employ DQN-based reinforcement learning to the DTI task to solve the cold-start problem by leveraging learned knowledge to transfer to the unknown drug-target interaction. Additionally, we utilize biological information from multiple sources, including sequence features and compound features, to reinforce the learned environmental information and improve the prediction accuracy. Experimental results on three benchmark datasets demonstrate that our model outperforms the baselines, validating the effectiveness of the proposed approach.