Algorithms and Software for Biomolecular Structure Determination at the Proteome Scale

Project Abstract Algorithms and Software for Biomolecular Structure Determination at the Proteome Scale One of the central tenets of structural biology is that the direct observation of molecules at the atomic level can reveal the function and mechanism of the biomolecular machines that are responsible for life. Recent advances in cryo-electron microscopy (cryo-EM) and tomography (cryo-ET) have revolutionized our ability to study these biomolecular machines, enabling the production of atomic resolution structures of large complexes and visualizing their native conformations and structural context in situ. These techniques hold great promise for resolving the conformational and structural heterogeneity intrinsic to the function of these biomolecules. However, the low throughput and limited scope of current methods hinder our ability to study the vast diversity of biomolecular structures and interactions across entire proteomes. In this proposal, we aim to develop novel artificial intelligence (AI) and machine learning (ML) methods to address critical challenges in cryo-EM and cryo-ET image processing. Our goal is to advance a new paradigm of structure determination that can operate at the proteome scale, enabling high-throughput and high-resolution analysis of complex biological systems. Specifically, we propose to integrate prior structural information from sequence and structural databases with cryo-EM and cryo-ET data, leveraging the strengths of both data-driven and knowledge-based approaches. By designing forward models that incorporate geometric information about the underlying protein structures and developing multimodal generative models that can generalize across diverse biomolecular complexes, we aim to significantly enhance the accuracy and efficiency of biomolecular structure determination. From this research, we will create open-source software tools and release curated datasets to support downstream tasks in machine learning for biology. The successful implementation of this proposal will facilitate the emergence of visual proteomics, transforming the current paradigm of reductive structural biology which focuses on individual targets. By enabling comprehensive analysis at the proteome scale, this work has the potential to fundamentally change our understanding of subcellular processes and lead to new discoveries in the molecular mechanisms underlying health and disease.