A pharmacophore is the three-dimensional representation of the features that characterize a receptor-ligand interaction. They can be used in virtual drug screening to identify ligands that match a particular set of interactions. They can be identified from a ligand alone, from a receptor-ligand complex, or from a receptor alone (apo approach).<br/><br/>We developed Phinder with the goal of semi-automating the apo approach to pharmacophore identification. Phinder docks nine small molecular fragments against an input receptor within a user-selected binding site. The docking is performed with Gnina, which uses Monte Carlo sampling to sample conformations of the receptor-fragment complex. The docked fragments are scored with an empirical scoring function, a convolutional neural network (CNN)-based affinity score, and a CNN-generated score indicating the probability that the pose is correct. They are then matched based on chemical structure to six pharmacophore feature types. Fragments matching a particular feature type are grouped via hierarchical clustering. The resulting clusters are characterized by their radius, the number of fragments overall and of each type that contributed, and the three scoring functions described above. To improve Phinder’s ability to identify true features, eight machine learning (ML) models were trained on a subset of the PDBBind Refined Set using Scikit Learn. Model performance was evaluated using area under the curve (AUC) for each pharmacophore type. The best performing models were trained on the entire PDBBind Refined Set using 5-fold cross validation. Top 10 analysis was used to determine Phinder’s ability to identify true features. We describe Phinder's success at recapitulating known pharmacophores derived from the protein-ligand structure and generating pharmacophores suitable for virtual screening.

Phinder: Semi-automated pharmacophore elucidation

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

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