Allie Getty, “Predicting the Binding Affinity of Protein-Protein Interactions”
Mentor: Arjun Saha, Chemistry & Biochemistry, Letters & Science (College of)
Poster #87
Our goal is to establish a reliable way of predicting binding affinity of protein-protein interactions (PPIs) using machine learning/AI. This would greatly improve the efficacy and efficiency of in silico drug design. Currently when designing novel peptide compounds, researchers have to test the binding affinity in a wet laboratory, which is expensive and time consuming. I have used 3 protein docking programs (LightDock, HawkDock, and HDock) and 1 protein structural prediction program (AlphaFold) to determine the binding affinity of 63 molecules. The binding affinity was determined using RMSD scores of the predicted model compared to the known X-ray crystallography structure. The predicted models are ranked by the programs. The best ranks do not always mean the best pose by RMSD score. I aimed to determine which programs give the best poses by RMSD score and which PPIs have a discrepancy between rank and RMSD score. By determining the difference in predicted affinity of known bonds, we can train our program to correctly predict the affinity of unknown interactions. The ability to predict the affinity of potential drug compounds digitally would speed up early stage drug development by weeding out molecules with low affinity.