Modern machine learning technologies has highly increased the scale of computer-aided drug design in drug discovery. Here, we applied advanced deep learning technology to create a deep convolutional generative adversarial network (DCGAN) named DRUGAN for the <i>de novo</i> design of novel agonists and antagonists of cannabinoid receptors, CB1 and CB2. DRUGGAN is a general computational framework which is readily to be applied in training DCGAN models for drug targets. It is composed of two deep neural networks, discriminator and generator. MACCS fingerprint was applied to represent the molecules with various structural features. The known CB1 or CB2 active compounds collected by ChEMBL belong to the “active” class, while inactive compounds belong to the “inactive “class. Randomly-selected ZINC compounds which have the same distributions of molecular weight and ClogP values are also in the “inactive” class. A DCNN architecture, AlexNet, was employed for the construction of both Discriminator and Generator neural networks. In the process of training, the two neural networks contested with each other in a zero-sum game, and were trained simultaneously. Two training strategies, full training and partial training were adopted. For the former, converted active and inactive compounds directly served as input for training; While for the latter, the maximal common substructure (MCS) of active compounds was reserved by fixing the corresponding MACCS feature bits to be 1 in the Generator-generated fingerprints. After training, similarity searches were performed to obtain the most similar real compounds in ZINC and ChEMBL databases using each generated fingerprint as the query. The identified real compounds were further evaluated using multiple modeling techniques including molecular docking and molecular dynamics simulations and binding free energy analysis. As partial training can generate structures with reserved MCS, this special type of DRUGGAN belongs to fragment-based drug design and can be applied to conduct drug lead optimization.

Drugan: A deep convolutional generative adversarial networks (DCGAN) for drug target-specific compound library construction

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.