LAUSR

Abstract Recurrent neural network (RNN) is one of the most representative architectures in deep learning and widely adopted in many research fields especially in natural language processing. In this work, a gated-recurrent-unit (GRU)-based deep neural network combined with transfer learning was successfully employed to establish a molecular generative model of ADAM10 inhibitors. The results showed that the GRU-based generative model can learn accurately the SMILES grammars of the molecules and be capable of generating novel potential ADAM10 inhibitors. In comparison with traditional ligand-based methods, the GRU-based generative model requires only the SMILES information of the chemical ligands and can generate efficiently a large set of potential novel structures. These unique advantages make it extremely useful in de novo drug design and large-scale virtual screening researches.