Automated identification of protein filaments in high-resolution 3D image data

COLLABORATORS: Dr Vicki Gold (CLES) and Dr Danielle Paul (Turing Fellow, University of Bristol)

Description: Cryo-electron tomography (cryoET) is a specialised 3D high-resolution imaging method used to visualise protein complexes and determine their structures. We image protein filaments, which are long hair-like assemblies that play important roles in a plethora of cellular processes. Examples include actin and myosin found in muscle, as well as pilins/flagellins which drive microbial motility.  By averaging many thousands of copies of protein filaments together from our 3D data, we can determine high-resolution protein structures at atomic-level detail (Fig. 1). Such structures help to reveal fundamental insight into their function, which provides key knowledge for healthcare and drug discovery programmes. A major limitation of our work is filament identification and selection for downstream processing, which is mostly a manual process. In this proposal, we aim to establish the feasibility of automated filament identification in cryoET data using two- and three-dimensional convolutional neural networks. Removing this restrictive bottleneck will change the way in which cryoET imaging data can be processed, increasing efficiency and increasing the accessibility of our method to the scientific community.