Data-driven biomolecular structure modeling for cryo-EM maps - Enter the text here that is the new abstract information for your application. This section must be no longer than 30 lines of text. Cryo-electron microscopy (cryo-EM) has become a widely used technique in structural biology for determining 3D structures of biological macromolecules. Despite an increasing number of structures being deposited in public databases like PDB and EMDB, many maps are still determined at 3 Å or worse resolution, posing challenges for structure modeling. Thus, there is a strong need for computational tools to assist in structure modeling and validation, given the increasing use of cryo-EM and cryo-electron tomography (cryo-ET) by scientists. Computational modeling is an integral and indispensable component in structural biology. Similar to the situation with microscope, superior modeling methods have the capacity to extract more accurate structural information from otherwise less informative data from cryo-EM and cryo-ET and provide new insights and opens up new research strategies. The goal of this project is to develop and apply computational methods for biomolecular structural modeling for cryo-electron microscopy. In this project I will substantially expand and enhance the capabilities of structure modeling methods to meet new demands and to improve accuracy and efficiency. This will be achieved by developing a deep learning-based approach that can consider key aspects in structure modeling altogether, including structure heterogeneity, atom detection in the density, structure prediction, interaction between proteins and nucleic acids. For lower-resolution maps, we will introduce a novel approach designed to identify and enhance key structural features within the map, which will significantly improve the accuracy of model building. This approach will also be applied to structure fitting and identification for cryo-ET. Additionally, a method for detecting and modeling small-molecule ligand structures in medium-high resolution EM maps will be developed, which is needed for drug discovery. In the process of building model structures, model validation is of crucial importance. To pinpoint modeling errors in PDB, we have devised a pioneering deep-learning-based quality assessment score, known as DAQ. We intend to expand DAQ's capabilities to identify atom-level errors in protein and nucleic acid structures, while also offering suggested corrections for flawed structure models within the DAQ-Score Database, which presently delivers model validation reports.
