Convolutional neural networks can successfully identify correct binding modes and meaningfully rank and score compounds given a protein-ligand complex. In our approach, the protein-ligand complex is discretized onto a 3D grid of atomic densities. The use of grids means that input of the convolutional neural network input directly maps onto interprettable spatial relationships. We will discuss a variety of methods in where grid-based convolutional neural network scoring functions can be used to <i>generate</i> new ligand poses and even new ligands. All code is available under an open-source license as part of the gnina project (https://github.com/gnina).

Generative models for structure-based drug design

256th National Meeting (2018)

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