In structure-based drug discovery, the energetic cost of replacing binding site water with a ligand could reveal significant clues about the chemical interactions that can take place. However, methodologies that utilize this information have not been fully evaluated, and in a field where more tools mean reaching solutions more efficiently, this is an important vacancy. One algorithm that calculates such energy terms is the inhomogeneous solvation theory. Its grid implementation, grid-based inhomogeneous salvation theory (GIST), which maps the resulting calculation onto a 3D grid, was found in a previous study to have the potential as a meaningful scoring metric in ligand-binding predictions. Our objective is to investigate if this algorithm will have a positive impact on predicting correct sets of chemical features that a binding ligand should possess and their respective spatial positions (known as pharmacophores), given a protein structure with a known binding site. To do this, we built a machine learning model that incorporates GIST grids, which was then trained and tested on molecular simulation data of solvated proteins from the PDBBind dataset. The model’s performance is evaluated based on prediction accuracies of a total of 13577 feature instances with one or more chemical interaction types from this dataset. Preliminary results with convolutional neural network implementation indicate that GIST raised overall prediction accuracies slightly, but its potential merits further investigation.

Evaluating a water displacement energy-based pharmacophore chemical feature prediction model

ACS Spring 2022

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