J. Med. Chem. | Release: a new drug design model for deep learning based on target structure

This paper introduces an article from the research group of Professor Hou tingjun of Zhejiang University 、 Professor Cao Dongsheng's research group of Central South University 、 A paper jointly published by the research group of Professor lihonglin of East China University of science and technology . This paper presents a method that can take protein into account in the process of molecular formation - Deep learning generation model of ligand interaction RELATION, This model is applicable to the new drug design based on target structure .RELATION The model uses both a million - magnitude molecular library and proteins - The ligand sets data to train the variational self encoder , After introducing two-way transfer learning , The sampling of the hidden layer can take into account both the novelty of the generated molecular skeleton fragment and the affinity to the target protein .RELATION The model also provides pharmacophore constraint generation and Bayesian optimization （BO） Sampling module , It can be customized for users to generate molecules with higher matching degree of pharmacophores and better docking scoring performance for targets .

Thesis link ：

https://pubs.acs.org/doi/10.1021/acs.jmedchem.2c00732

RELATION Model framework of method