Description
Date depot: 16 octobre 2023
Titre: Deep learning methods for analyzing continuous conformational variability of biomolecular complexes from cryo electron tomography data
Directrice de thèse:
Slavica JONIC (IMPMC (EDITE))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Intelligence artificielle
Resumé: Cryo electron tomography (cryo-ET) allows studying the structure and dynamics of biomolecular complexes in their cellular environment. A thin cell section is tilted in the microscope and an image is collected for each tilt angle. A 3D reconstruction (tomogram) is calculated from the tilt series and multiple instances of the same target biomolecular complex are extracted from the tomogram into individual volumes (subtomograms). The subtomograms are difficult to analyze because of their low signal-to-noise ratio and their deformations induced by a limited tilt angle of the data collection (the so-called missing wedge). The standard approach to analyze the subtomograms is their iterative alignment and classification into a small number of classes and the class averaging, which increases the signal-to-noise ratio of the reconstructed average conformation of the complex but hides slightly different individual conformations. The identification of the entire conformational variability of the complexes is important for understanding the molecular mechanisms of the complexes. We have previously developed HEMNMA-3D [1], TomoFlow [2], and MDTOMO [3], which are the first three methods for analyzing continuous conformational variability of biomolecular complexes from cryo-ET data and obtaining the entire conformational landscape. These methods use classical image analysis in combination with motion field modelling using either molecular dynamics simulations (HEMNMA-3D, MDTOMO) or optical flows (TomoFlow). These methods have a high computational cost when analyzing large sets of subtomograms. This PhD thesis will focus on the development of deep learning methods for analyzing continuous conformational variability of biomolecular complexes from cryo-ET subtomograms, to allow analyzing large amounts of subtomograms. Both supervised and unsupervised deep learning methods will be explored in this context. Also, in this PhD thesis, the candidate will investigate the potential of the use of deep learning for the missing wedge correction and the potential of analyzing challenging tilt sub-images, which are very noisy but not affected by the missing wedge, contrary to subtomograms. The new methods will be validated with synthetic and experimental data, in collaboration with biologists at the IGBMC, Illkirch. This thesis is part of a collaborative, interdisciplinary research project funded by the ANR.[1] Harastani, M., Eltsov, M., Leforestier, A., and Jonic, S. (2021). HEMNMA-3D: Cryo Electron Tomography Method Based on Normal Mode Analysis to Study Continuous Conformational Variability of Macromolecular Complexes. Front Mol Biosci 8, 663121. Open-access: https://doi.org/10.3389/fmolb.2021.663121.[2] Harastani, M., Eltsov, M., Leforestier, A., and Jonic, S. (2022). TomoFlow: Analysis of Continuous Conformational Variability of Macromolecules in Cryogenic Subtomograms based on 3D Dense Optical Flow. J Mol Biol 434, 167381. https://doi.org/10.1016/j.jmb.2021.167381; https://hal.science/hal-03452809.[3] Vuillemot, R., Rouiller, I., and Jonic, S. (2023). MDTOMO method for continuous conformational variability analysis in cryo electron subtomograms based on molecular dynamics simulations. Sci Rep 13, 10596. Open-access: https://doi.org/10.1038/s41598-023-37037-9.
Doctorant.e: Feyzi Florène