We describe a convolutional neural network (CNN) deep learning scoring function for structure-based design that takes as input a comprehensive 3D representation of a protein-ligand interaction. A CNN scoring function automatically learns the key features of protein-ligand interactions that determine binding. We train and optimize our CNN scoring functions to discriminate between correct and incorrect binding poses as well as known binders and non-binders. We describe how the output of the CNN model can be meaningfully visualized at an atomic granularity. We find that our CNN scoring function outperforms the AutoDock Vina scoring function when ranking poses both for pose prediction and virtual screening. Current limitations of the approach and potential future directions will be discussed.

Protein-ligand scoring with convolutional neural networks

253rd National Meeting (2017)

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