Understanding the three-dimensional structure of molecules is a key part of understanding their function. Similarly, our ability to engineer new materials to meet advanced functional requirements, is founded on our understanding of molecular structure. In the field of molecular biology, the determination of molecular structure requires us to be able to determine 3D properties from images.
Explaining the science
Cryogenic electron microscopy (cryo-EM) is an imaging method that enables the determination of biomolecular structures at the atomic level. One of the most important outstanding problems in the field is that of reconstructing cryo-EM data sets in which the projection images come from an ensemble of different structures with continuously changing conformations. Hidden within this complexity are the essential functional dynamics of the system. However, reconstructing molecular structure and dynamics from rich but noisy images and volumes remains a challenging problem.
In this project, we will develop new methods, involving expertise in data science, applied mathematics, computer vision, machine learning and inverse problems, to address the challenges. The first major challenge is reconstructing molecular structure from the observations; to attack this problem, we will leverage our expertise in joint reconstruction and deformation models. The second major challenge is incorporating physical constraints as a prior during the determination of molecular structure. Adding methods to cope with dynamic and continuously changing structures would be a considerable advance. Finally, we will develop approaches that will allow us to intelligently identify candidate structures within images or volumes, and tools to generate the large amounts of annotated training data that will be required for the project.
To address this challenge, we have assembled an interdisciplinary team with groups at the Turing, University of Cambridge (Department of Applied Mathematics and Theoretical Physics, DAMTP), MRC Laboratory of Molecular Biology (MRC-LMB) and the Science and Technologies Facilities Council (STFC).
Development of computer vision and machine learning approaches to enable the intelligent sampling of planar and volumetric cryo-EM data. These approaches may be informed by prior knowledge and/or physical models of molecular structure.
Develop a software framework to record metadata associated with image processing, reconstruction and atomic modelling and simulation to enable next generation ML algorithms; apply experimentally determined molecular motions to atomic models [Link coming soon]