Enhancing the resolvability of cryo-EM maps in protein-ligand complexes using deep learning
Abstract
Understanding protein-ligand interactions at the atomic level is crucial for unraveling how ligands influence macromolecular functions. This insight can be applied to the development of pharmaceuticals through structure-based drug discovery. However, traditional X-ray crys- tallographic methods for obtaining experimental structures of such complexes are challenging and time-intensive. Recent advances in single-particle cryo-EM have addressed this challenge, enabling the determination of atomic-resolution structures of complex biomolecular systems. While cryo-EM now provides exceptional resolution for overall structures, ligand resolutions remain too low for accurate modeling. Recently, artificial intelligence (AI)- based methods have been developed to model and refine structures based on EM data; however, their primary focus has been on protein accuracy. Here, we exploited the increased power of data-driven research and building an AI model based on vision transformer to improve low-resolution ligand maps and refine structural models for small molecules in protein-ligand complexes.
