Enhancing structure-based drug‒target interaction prediction via electron-density map-informed conformation optimization

The identification of drug‒target interactions is a critical in early drug discovery but remains challenging due to high attrition rates and substantial costs. Structure-based prediction methods such as molecular docking are widely used; however, their performance highly depends on protein structure quality. We hypothesize that optimization of protein structures with electron density maps and conformation ensembles can improve the performance of structure-based drug‒target interaction prediction methods. To this end, we developed ElectMap, a protein structure optimization method that integrates molecular dynamics simulation data with cryo-EM density maps through a 3D U-Net-based framework. ElectMap processes each protein individually, generating an optimized conformation based on its own simulation conformation ensemble to fit the experimental density map, and providing an optimized structure for drug‒target interaction prediction. Across representative protein target classes (G-protein coupled receptors, ion channels, transporters, catalytic receptors, and enzymes), ElectMap consistently improved the correlation between predicted binding affinities and experimental bioactivities for physics-based molecular docking, and yield target-dependent improvements for machine-learning-based approaches. ElectMap emerges as a powerful structure optimization tool for enhancing structure-based drug discovery.