Predicting Ligand Binding Modes by Scaffold-Guided Structure Refinement

Abstract

Efficient structure-based drug design relies on knowledge of a ligand’s binding pose and its specific interactionsinformation that is often not available experimentally. Despite the plethora of binding mode prediction methodsincluding cofoldingachieved accuracies are often insufficient. Here, we present “Scaffold-Guided Structure Refinement”, leveraging information on known binders within ligand series targeting a specific protein. Our method is based on the observation that shared molecular scaffolds among binders exhibit conserved binding modes. By applying molecular docking to diverse target model conformations, we identify those simultaneously allowing consistent scaffold placement, favorable interactions and low ligand strain. We demonstrate this approach’s ability to optimize models from different initial sourcesincluding an inaccurate cofolding modelin three case studies. In all cases, we successfully identified critical induced fit effects and accurately reconstructed near-native ligand binding modes with scaffold root-mean-square deviation (RMSD) values of at most 2.2 Å.