Multi-objective de novo Drug Design using Bidirectional Recurrent Neural Networks and Nondominated Sorting

Drug Design (DD) is a significant field for pharmaceutical businesses study. Deep Learning (DL) approaches are evolving as an auspicious solution to employ de novo drug design, such as to develop drug-like molecules personalized to particular requirements. In DD, the design as well as production of safe and effective compounds is a long-term, complex, as well as difficult process. Hence, this research proposes the Bidirectional Recurrent Neural Network (BiRNN) for the design of multi-objective de novo drug. Initially, the CheMBL and KIBA datasets are utilized to estimate the effectiveness of the proposed method. Then, the data are proprocessed by utilizing Fragment similarity calculations to design various ligands sets and particular target collections. The proposed BiRNN attains the accuracy of 0.9742, and 0.0632 on CheMBL dataset, as well as Mean Square Error (MSE) of 0.213 and 0.024 on KIBA dataset respectively. These obtained results are better and effective when compared to the existing methods like Transformer-Decoder Generator, A2C Reinforcement Learning (RL) and Recurrent Neural Network (RNN).