Comparison of Traditional and Machine Learning Programs in the Evaluation of Protein-Ligand Binding


Authors

Alexander Zatuchney, Blessing Anyangwe, Arushi Desai, Elizabeth Fishman, Kevin Jin, Kai Kim, Erin Kraus, Eugene Lee, Angelina Li, Bridget Liu, Nicholas Sardy, Aarna Tekriwal, Osariemen Unuigbe, Eric Zhu, and David Cincotta, Drew University, United States of America

Abstract

Molecular docking, an in-silico method with widespread pharmacological applications, is used to predict the optimal conformation of a protein-ligand complex. Traditionally, it uses search-score algorithms that generate protein-ligand poses and calculate each pose’s binding strength. More recently, artificial intelligence (AI) programs have been developed and trained with protein-ligand datasets. To compare the accuracy of these approaches in site-specific docking, a traditional program and a deep-learning (DL) program were tasked with docking a set of protein-ligand pairs. Upon comparison of the two programs’ results, it was determined that they predict optimal binding conformations with similar accuracy.

Keywords

Molecular docking, Protein-ligand binding, Deep-learning, DiffDock-Pocket, SeeSAR