![]() None of these approaches, however, have been able to demonstrate generalization, meaning that for each protein of interest, users must first manually annotate hundreds to thousands of particles in tomograms and train the neural network to identify that protein. This has led to the development of several deep learning-based tools often leveraging popular 3D-Unet convolutional neural network (CNN) architectures 19, 20, 21. The accurate localization of macromolecules inside cryo-electron tomograms is a well-recognized barrier for studying cellular life at the mesoscopic level 18. To perform STA, however, particles of a macromolecule of interest must first be located within the tomograms, a task complicated by the three-dimensional (3D) nature of these data. Particularly when complemented by recent advances in structure prediction such as alphafold2, STA forms a powerful crossbridge between protein biochemistry and cellular proteomics 15, 16, 17. Cryo-ET offers a unique opportunity to capture cellular processes in three dimensions and in unprecedented detail, and subsequent analysis of specific macromolecules from tomograms through subtomogram averaging (STA) allows for in-depth structural determination of macromolecular complexes in situ 11, 12, 13, 14. Advances in high-pressure freezing and focused ion beam milling at cryogenic temperatures now allow for the routine preparation of thin (less than 200 nm) lamellae from cells or even small organisms 8, 9, 10. By embedding tomograms in an information-rich, high-dimensional space that separates macromolecules according to their three-dimensional structure, TomoTwin allows users to identify proteins in tomograms de novo without manually creating training data or retraining the network to locate new proteins.Ĭryogenic-electron tomography (cryo-ET) has emerged as a landmark technique for the visualization of macromolecules within their native cellular environment 1, 2, 3, 4, 5, 6, 7. To assist in this crucial particle picking step, we present TomoTwin: an open source general picking model for cryogenic-electron tomograms based on deep metric learning. Available methods for this task suffer either from being error prone or requiring manual annotation of training data. Detailed analysis of macromolecules through subtomogram averaging requires particles to first be localized within the tomogram volume, a task complicated by several factors including a low signal to noise ratio and crowding of the cellular space. Cryogenic-electron tomography enables the visualization of cellular environments in extreme detail, however, tools to analyze the full amount of information contained within these densely packed volumes are still needed.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |