Description
Date depot: 22 novembre 2021
Titre: Geometric deep manifold learning combined with natural language processing for protein movies
Directrice de thèse:
Elodie LAINE (LCQB)
Encadrant :
Sergei GRUDININ (LJK)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Intelligence artificielle
Resumé: The goal of this project is to explore the contribution of recent methods of statistical learning and deep neural networks to predict motions and conformational states relevant to protein functioning. In other words, we aim at learning low-dimensional motion manifolds using sparse high-dimensional observations (3D structures). Our specific objectives will be to:
1. Develop algorithms capable of operating on compact representations of geometric data structures, taking into account the specific (physico-chemical) constraints applying on them and the uncertainties in the observations;
2. Develop a generative model able to recover observed states, interpolate between them and extrapolate to previously unseen states. The generation will be conditioned on the protein sequence of amino acids, and on low-resolution experimental data for guiding the extrapolation;
3. Generate new plausible states for a set of proteins with therapeutic interest, that could be targeted by small molecules toward modulating their function.
Doctorant.e: Lombard Valentin Luka