A MODEL FUSION BASED REINFORCEMENT LEARNING METHOD FOR DRUG-TARGET INTERACTION PREDICTION

In order to address the issue that the performance differences between the machine learning models lead to inaccurate prediction of DTI on different datasets, a model fusion based reinforcement learning method for DTI is proposed(MF-RLDTI). We trained the three models, CMF, WNN-GIP, and NetLapRLS, and extracted the prediction score matrices separately. The Sarsa algorithm was used to continuously optimize the weights of the three matrices. The linear weighting was performed to output the final prediction result. Compared with different prediction models, the experimental results show that MF-RLDTI has high accuracy in DTI prediction.