Modeling Receptor-ligand interactions for Drug Discoveryfrom Cryo-EM data using AI

Project Summary Cryo-electron microscopy (cryo-EM) has become a widely used experimental technique for determining three- dimensional (3D) structures for biological macromolecules. Its impact is not confined to academic research alone; biotech and pharmaceutical companies have increasingly adopted cryo-EM for its ability to provide detailed structural insights into biological targets. When the resolution of cryo-EM density data is at 3 Å or higher, drug molecules and their interactions with target biomolecules can be directly visualized within cryo-EM density maps. In the pharmaceutical industry, cryo-EM is particularly transformative for drug discovery, as it allows for the visualization of drug-target interactions in their native states, facilitating the identification and optimization of new therapeutic candidates. However, a challenge in using cryo-EM for drug discovery is that achieving a high resolution better than 3 Å is not always guaranteed. When the resolution is worse than 3 Å, ligands may still be visible, but the process of modeling becomes considerably more time-consuming and error-prone. This underscores the need for advanced software tools that can streamline the modeling process, reduce errors, and make cryo-EM more accessible to non-specialists in drug discovery efforts. The goal of this project is to develop and provide state-of-the-art structure modeling and ligand modeling software capable of accurately modeling bound drugs in receptor structures from cryo-EM data with resolutions up to around 5-6 Å. This technology can accelerate the development of novel drugs by offering precise structural information that can guide the design of molecules with improved efficacy and reduced off-target effects, ultimately advancing the field of precision medicine. This Small Business Innovation Research Phase I project aims to expand and advance structural modeling and analysis for drug discovery using cryo-EM by using state-of-the-art deep learning techniques. Our key innovations include: (1) To improve receptor protein modeling accuracy using advanced deep learning techniques; (2) and by considering conformational heterogeneity in cryo-EM data; (3) ligand detection and modeling in cryo-EM maps using deep learning and physics-based modeling and refinement; (4) The seamless automated pipeline for modeling protein, nucleic acids, and bound ligand molecules. The intellectual merit of this project lies in its methodology, which overcomes the current limitations in biomolecular modeling for cryo-EM data.