Description
Date depot: 2 avril 2023
Titre: Unsupervised deep learning approaches to extract atomic-scale conformational landscapes of biomolecules from cryo electron microscopy images
Directrice de thèse:
Slavica JONIC (IMPMC (EDITE))
Directrice de thèse:
Catherine VÉNIEN-BRYAN (IMPMC (ED515))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Intelligence artificielle
Resumé: Structural studies of biomolecular complexes are essential for understanding their working mechanisms. Cryo electron microscopy (cryo-EM) is the mainstream structural biology technique that allows simultaneous determination of multiple conformational states of the same complex. Moreover, cryo-EM has a potential to obtain information about the full conformational variability of a complex (conformational landscape) and continuous conformational transitions of complexes (gradual transitions with many intermediate states instead of a few discrete states). However, image analysis methods should be developed for accurate and fast extraction of such information. This PhD thesis will be focused on the development of unsupervised deep learning methods to extract atomic-scale conformational landscapes of biomolecules from cryo-EM images (supervisor: SJ). The new methods will be tested not only using typical test data of large complexes, available in the public data archive EMPIAR (e.g., ribosomal complexes), but also using data of more challenging, small complexes, such as membrane proteins. Cryo-EM images of membrane proteins are difficult to analyze because of their small size, a low signal-to-noise ratio (due to data collection using low electron dose to minimize sample radiation damage), and occlusions (due to surfactants, such as detergent, that are used to extract proteins from the membrane and purify them in their native state, which hide the protein in the image and may affect its dynamics). In this thesis, the new methods will be tested using cryo-EM images of the human potassium channel Kir2.1, which are already available (supervisor: CVB). Kir2.1 is a membrane protein with multiple functions (e.g., regulating pancreatic insulin secretion, controlling muscle contraction, regulating pacing in cardiac cells and neurons, etc.) and its disfunction is responsible for a number of human diseases (e.g., Andersen’s syndrome, Bartter’s syndrome, neonatal diabetes, etc.). The new methods will help analyze the dynamics of Kir2.1, which will in turn help better understand its function and dysfunction. Each supervisor has unique expertise in France: development of innovative cryo-EM image analysis methods (SJ, ED 130) and functional and structural studies of potassium channels at high resolution using cryo-EM (CVB, ED 515).