Efficient discovery and production of new molecules are essential and understanding protein-ligand interactions is a critical step in facilitating small molecule design. Accurate prediction of protein-ligand interactions can elucidate various biophysical and biochemical properties from substrate binding to therapeutic design application. Currently, experimental characterization of protein-ligand binding affinities is the most accurate method, however, this is time-consuming and labor intensive. Implementing physics-based computational methods and developing machine learning models to investigate such protein-ligand interactions are crucial yet challenging due to the large area of chemical space necessary to explore. We recently introduce the expanse of a novel, distance-aware, machine learning approach to integrate knowledge representation and reasoning for the prediction of protein-ligand interactions. Atomic-level information of the interacting ligand and protein is directly embedded into a graph-based structure to mimic the interaction’s actual environment. Through experimentation, we have observed the best performance from a model that uses distinctly engineered features for the protein and ligand, and promising results from a similar model that does not require the prior knowledge of intermolecular interactions.

Decoding protein-ligand interaction through graph neural networks

ACS Spring Meeting 2021

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