Exploration et de la modélisation structurale d’interactions protéiques guidées par l’information évolutive

Informations générales
Nom
Quignot
Prénom
Chloé
Diplôme
Thèse
Année
2020
Détails de la thèse/HDR
Université
Jury
Isabelle André, Directeur de Recherche, CNRS, INSA - rapporteur
Jean-Christophe Gelly, Maître de Conférence, Université de Paris - rapporteur
Pablo Chacon, Directeur de Recherche, Institute of Physical Chemistry « Rocasolano », Espagne - examinateur
Annick Dejaegere, Professeur, Université de Strasbourg - examinatrice
Olivier Lespinet, Professeur, Université Paris-Saclay - examinateur
Raphaël Guerois, Directeur de Recherche, CEA Saclay - directeur de thèse
Jessica Andreani, Chargée de Recherche, CEA Saclay - co-encadrante de thèse
Directeur (pour les thèses)
Raphaël Guerois
Jessica Andreani
Résumé en anglais
Protein complexes are of fundamental importance in most biological processes and mainly carry out their function in networks. The structure of their interface can give us crucial information to understand the mechanisms behind these processes. This thesis focuses on the improvement of the performance of structural prediction methods, in particular by exploiting co-evolutionary information.

As part of my PhD project, I participated in major developments in our docking server, InterEvDock2, which suggests 10 interface models for a pair of input proteins using a mix of different scoring properties. InterEvDock2 now also accepts oligomeric structure inputs or sequence inputs, for which it can model monomeric structures, as well as user constraints taken from prior knowledge of the interaction. I validated the performance of InterEvDock2 on a large benchmark of 812 docking cases and found that InterEvDock2 was capable of finding a correct complex structure in as much as 32 % of these cases. My work then focused on finding a more efficient and explicit way of integrating implicitly defined evolutionary information into scoring. I made this information directly compatible with atomic-scale scoring thanks to homologous interface modelling. This strongly increases predictive power, from 32% to 40% on a large benchmark. Moreover, throughout my PhD, I was able to participate in 10 blind-test docking challenges through CAPRI (Critical Assessment of Predicted Interactions). The strategies applied by our team, which enabled us to rank first in the latest CAPRI round for 2016-2019, are described in the last chapter of this manuscript. This work aims at helping biologists study their proteins or biological pathways of interest using well performing prediction methods. It constitutes a step towards the final goal of interactome prediction.