Molecular generative models using 3D information of target pocket are recently gaining increasing attention in the sector of <i>de novo</i> drug design. Given the abundance of high-quality X-ray diffraction experiment data in PDB and the fact that electron density reflects the nature of the atoms in terms of physical and chemical properties, we have developed a molecular generative model using electron density as pocket representations. The model is built on top of 3DCNN and GAN, and trained based on over 120,000 experimental electron density maps from PDB. Taking the coordinates of a target pocket or its electron density map as input, the model can generate small molecule binders with their favorable binding conformations. We hereby present two cases demonstrating the advantage of using our model for <i>de novo</i> drug design. In the first case, by analyzing non-covalent interaction between generated compound and the pocket, we show that our model provides high fulfillment of the requirements for the favorite binding mode. In the second case, by comparing with the classical virtual screening against commercially available library and that constructed by 2D molecular generative model, our model shows improved enrichment of active compound.

From target pocket to high quality drug-like molecules with favorable binding conformation: the application of a novel pocket-based 3D molecule generation model

ACS Fall 2021

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