Ensemble docking (ED) is a widely used approach to incorporating protein flexibility during structure-based virtual screening (SBVS) campaigns. ED consists of docking the ligands to a set of different protein conformations, aiming to represent the structural diversity of the target’s binding site. Although ED has been successfully applied in several studies, critical challenges still need to be addressed. One open question is how to identify the most druggable protein conformations; those that will perform the best in a virtual screening experiment. Here, we implemented a convolutional neural network (CNN) model to predict the druggability of a given protein conformation, using the structural information of its binding site. For this purpose, we selected the CDK2 protein as a case study, as it has been widely employed to test and validate SBVS methodologies. Conformational ensembles were built from X-ray crystallographic structures, and molecular dynamics (MD) derived conformations. Then, a subset of these conformations was used to perform ensemble docking assays along with molecules from the DEKOIS 2.0 benchmarking set. After, we implemented the CNN using the docking performance of each protein conformation as the response variable, while the geometric and physicochemical features were used as predictive variables. Finally, we applied feature selection methodologies to identify the most relevant structural properties that could determine the effectiveness of a CDK2 conformation in an SBVS experiment.

Druggability prediction of protein conformations using convolutional neural networks

@LatinXChem is a virtual forum through which the community of Latin American chemists, located anywhere in the world, can share and discuss their research results and advances. More than 1200 posters from 43 countries were received in its first version, with 20 million impacts on Twitter. The second event organized by @LatinXChem will be the #LXChem2021 Conference (September 20, 2021), where posters divided into eleven categories can be discussed via Twitter.

LatinXChem 2021

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