Currently, machine learning approaches have had significant success in predicting protein-ligand binding affinities. However, recent papers have shown that there is significant bias in some commonly used benchmarking datasets, resulting in overly optimistic results of the models. Additionally, training data generation is typically done through redocking, which is not analagous to the use-case in drug discovery as one is trying to fit either similar molecules to known binders or fit known binders into new targets. Lastly, there is a lack of available, high quality, and labelled data to train our models on. In order to alleviate these concerns we present an extension of GNINA, a grid-based convolutional neural network, that is trained on a new set of cross-docked poses in order to generate a more reliable model. Cross-Docking provides both a method of data augmentation and a way to generate contexts for training data that is more similar to the use-case of drug discovery, by grouping the PDB by binding pockets as defined by Pocketome and docking every ligand with every receptor in a given pocket. We show that the performance of the new data is similar to training on clustered cross-validated splits of the PDBbind database, can generalize to the PDBbind models, and that models trained on redocking with the PDBbind data fail to generalize to cross-docked poses. We also explore the extent that these structure-based models actually use protein-ligand interactions to make their predictions versus learning ligand-only cheminformatic models.

GNINA 1.0: Predicting protein-ligand binding affinity through cross-docking

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

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