A Targeted Drug Design Method Based on GRU and TopP Sampling Strategies

Abstract

Deep learning algorithms can be used to improve the efficiency of drug design, which is a very meaningful research topic. This paper proposes a targeted drug design model based on the gated recurrent unit (GRU) neural network algorithm, which trains a large number of drug molecules obtained from the Chembl database for generating a generic and unbiased molecular library. For improving the efficiency and accuracy of the trained model, a fine-tuning strategy is used to train against the active compounds of the target protein. In addition, a TopP sampling strategy is used to sample molecular tokens for reducing the number of generated drug molecules that are invalid or existing drug molecules. Finally, the novel coronavirus 3CLpro protease is selected for verifying the effectiveness of the proposed model. Molecular docking results show that the molecules generated by the proposed model have lower average binding energies than the existing active compounds.