PhD Position F/M PhD position F/M : Machine learning for 3D cryo-electron

14 Feb 2024
Job Information

Organisation/Company

Inria

Research Field

Computer science

Researcher Profile

Recognised Researcher (R2)

Country

France

Application Deadline

14 Mar 2024 - 00:00 (UTC)

Type of Contract

To be defined

Job Status

Full-time

Hours Per Week

To be defined

Is the job funded through the EU Research Framework Programme?

Not funded by an EU programme

Is the Job related to staff position within a Research Infrastructure?

No

Offer Description

A propos du centre ou de la direction fonctionnelle

The Inria Rennes - Bretagne Atlantique Centre is one of Inria's eight centres and has more than thirty research teams. The Inria Center is a major and recognized player in the field of digital sciences. It is at the heart of a rich R&D and innovation ecosystem: highly innovative PMEs, large industrial groups, competitiveness clusters, research and higher education players, laboratories of excellence, technological research institute, etc.

Contexte et atouts du poste

The thesis will take place in the SAIRPICO project-team, which is specialized in the development of innovative methods for image restoration/reconstruction, motion analysis and computation of molecule trajectories in live cell imaging, and biophysical parameter estimation. The thesis we propose is at the frontier of applied mathematics, image processing/analysis, machine learning, and computer science. The goal is to develop generic image analysis algorithms for the analysis of 3D cryo-eletron tomograms. In particular, we will investigate innovative deep learning approaches (e.g., convolutional neural networks, transformers, Generative methods) to localize and identify macromolecules with a focus on nucleosome and linkers DNA in cell nuclei. From a methodological point of a view, we will focus on hybrid methods that bring together deep learning approaches based human-expert annotations and simulated annotations. The thesis will be supervised by Charles Kervrann (Inria Rennes).

Mission confiée

CONTEXT AND MOTIVATION - Our main goal is to reveal chromatin reorganization during genotoxic stress with unprecedented resolution and analysis detail, which will be allowed by combining cryo-ET with using new deep learning approaches that will be developed. The analysis of chromatin at three structural levels - organization of chromatin domains, local geometry of nucleosome fibers, conformation and disassembly of nucleosomes - will reveal the chromatin structure-based mechanisms enabling detection and repair on UV-induced lesions in the chromatin context.

In situ cryo-eletron tomography (cryo-ET) provides the ultimate quality of cell structural data available to date, because (i) biological imaged macromolecules without any chemical treatment or stain; (i) the sampling and information quality enables structure interpretation on the near atomic scale, (iii) biological molecules are captured in their native functional environment with interaction partners, preserving their native conformational space. At the same time, in situ cryo-ET data are challenging for information extraction because of (i) a signal-to-noise ratio (SNR) that is much lower than in other imaging applications, (ii) a high crowding of molecular components in cells, (iii) incomplete angular sampling during image tilt series collection inducing missing data in 3D Fourier space and associated deformations in 3D real-space reconstruction (the so-called missing wedge (MV) problem [1, 2]
THESIS OBJECTIVE -Developing generative models and supervised DL methods for spatial localization, identification, and spatial organization of nucleosomes in situ, and simulation of cryo-ET tomograms. Automated identification of macromolecules inside cellular tomograms is challenged by noise and reconstruction artifacts, as well as the presence of many molecular species in the crowded volumes. In past works, the SERPICO team investigated supervised deep-learning and 3D CNN (DeepFinder [3]) to accurately and simultaneously localize multiple macromolecules (e.g., ribosomes, proteasomes, ...) in 3D cell cryo-electron tomography images. Our results demonstrated that DeepFinder outperforms the usual template matching algorithms and most of competitive recent deep learning methods (SHREC 2019, 2020, 2021). On cellular cryo-ET data, DeepFinder localized membrane-bound and cytosolic ribosomes (3.2MDa), Rubisco (560 kDa soluble complex), and photosystem II (550 kDa membrane complex) with an accuracy comparable to expert-supervised ground-truth annotations. In collaboration with IGBMC Strasbourg (2023), DeepFinder was trained and adapted to localize nucleosomes in denoised 3D cryo-ET images (manuscript in preparation). Nevertheless, several thousands of annotations were supplied by the experts and correspond to 3D coordinates of manually labeled macromolecules. As the number of available cellular tomograms may be limited, annotations are very expensive and represent a hard task in 3D cryo-ET, and adaptation to any microscopy set-up is not well controlled with annotated data, we will focus in this thesis on unsupervised or weakly supervised methods [4] including low-dimensional representations and simulations.

BIBLIOGRAPHY -

Guesdon, S. Blestel, C. Kervrann, D. Chrétien. Single versus dual-axis cryo-electron tomography of microtubules assembled in vitro: Limits and perspectives,Journal of Structural Biology, 181(2):169-178, Feb. 2013

Moebel, C. Kervrann. A Monte Carlo framework for missing wedge restoration and noise removal in cryo-electron tomography, Journal of Structural Biology: X, 4:100013, pp. 1-18, doi: 10.1016/j.yjsbx.2019.100013 , HAL-INRIA-02424804 , 2020

Moebel, A. Martinez-Sanchez, L. Lamm, R.D. Righetto, W. Wietrzynski, S. Albert, D. Larivière, E. Fourmentin, S. Pfeffer, J. Ortiz, W. Baumeister, T. Peng, B.D. Engel, 10, C. Kervrann. Deep learning improves macromolecule identification in 3D cellular cryo-electron tomograms, bioRxiv , Nature Methods, 18: 1386-1394, HAL-INRIA-03509223 , doi: 10.1038/s41592-021-01275-4 , 2021

Moebel, C. Kervrann. Towards unsupervised classification of macromolecular complexes in cryo electron tomography: challenges and opportunities, Computer Methods and Programs in Medicine, Special issue on "Computational Methods for Three-Dimensional Electron Microscopy (3DEM)", bioRxiv-2022.03.10.483789 , doi: 10.1016/j.cmpb.2022.107017 , 2022

Herbreteau, E. Moebel, C. Kervrann. Normalization-equivariant neural networks with application to image denoising, arXiv-2306.05037 , doi: 10.48550/arXiv.2306.05037, 2023

Gubins et al. Classification in cryo-electron tomograms. In Proc. Eurographics Workshop on 3D Object Retrieval - SHREC - 3D Shape Retrieval Contest (2021), Vienna, Austria, May 2021

Compétences

Technical skills and level required :

Languages :

Relational skills :

Other valued appreciated :

Avantages

• Subsidized meals
• Partial reimbursement of public transport costs
• Possibility of teleworking (90 days per year) and flexible organization of working hours
• Partial payment of insurance costs

Rémunération

Monthly gross salary amounting to 2100 euros for the first and second years and 2158 euros for the third year

Requirements
Additional Information
Work Location(s)

Number of offers available

1

Company/Institute

Inria

Country

France

Geofield

Where to apply

Website

https://illbeback.ai/job/phd-position-f-m-phd-position-f-m-machine-learning-for…

STATUS: EXPIRED