Skip to Main content Skip to Navigation

Development and assessment of protein loop modeling methods: Application to CDR loops in antibodies

Amélie Barozet 1
1 LAAS-RIS - Équipe Robotique et InteractionS
LAAS - Laboratoire d'analyse et d'architecture des systèmes
Abstract : This thesis deals with antibody modeling, in particular the modeling of hypervariable loops found at the interface with the antigen. These protein loops are responsible for specific recognition of the antigen, as well as the formation of the antibody-antigen complex with a great affinity, possible thanks to a great variability of sequence and to the plasticity of these protein fragments. Indeed, contrary to other more stable structural elements like alpha helices or beta sheets, protein coils exhibit a flexibility that plays a crucial role in many biological processes. This manuscript starts with describing the analysis of structural changes in antibodies upon antigen binding, which constitutes the first contribution of this PhD research work. This study, based on the analysis of experimental structural data, shows that antibody conformational changes (occurring mainly in the loops), can be substantial and are not sufficiently accounted for. In particular, docking algorithms show poor results when dealing with excessively flexible hypervariable loops in the antigen binding site. In this context, the PhD research work then focused more generally on protein loop modeling. These flexible protein regions represent a challenge for structural biology. Most experimental data related to protein structure are obtained through X-ray crystallography, which cannot correctly represent flexible parts in the structure. Indeed, it provides a unique structure, which is inappropriate for protein loops, that adopt an ensemble of different conformations with various associated probabilities. As shown by multiple recent works, current protein methods cannot properly model protein loops. Protein loop modeling is usually performed in two steps. First, an exhaustive cong, and consists in attributing scores to each of these sampled conformations. This score is meant to represent the energy differences between the models generated during the first step. Sampling and scoring remain open problems. Indeed, methods developed so far in the field mostly focus on predicting a single stable conformation, that is not representative enough. The two next contributions of the PhD research work logically follow from this observation. The first one presents a method for exhaustive sampling, with a reinforcement learning component to speed up the generation of loop models. This robotics-inspired method uses a geometric representation that forbids steric clashes between atoms and uses protein fragments from a database built specially for this application. The second contribution is an in-depth analysis of the performance of several scoring methods on a set of flexible loops for which experimental data exist. Combining sampling and scoring allows the visualization of energy landscapes implicitly modeled by these methods. The analysis of these energy landscapes enables to precisely identify both the flaws of sampling and the limits of scoring methods. Finally, these methods were applied to an antibody with a hypervariable loop which changes conformations upon antigen binding. Results show that the methods previously studied and developed enable to model a consistent energy landscape for this flexible loop, identifying both known conformations. This suggests that these methods could be successfully applied to antibody design by predicting a loop's stability in a position or another and discarding loop sequences that are insufficiently stable or that adopt undesirable conformations. Although applied to antibodies, the research contributions presented in this work can perfectly be generalized to the analysis of protein loops in other systems, since the developed methods are not antibody-specific.
Document type :
Complete list of metadata

Cited literature [197 references]  Display  Hide  Download
Contributor : Laas Hal-Laas <>
Submitted on : Wednesday, January 15, 2020 - 3:01:28 PM
Last modification on : Thursday, June 10, 2021 - 3:06:38 AM
Long-term archiving on: : Thursday, April 16, 2020 - 3:58:57 PM


BAROZET Amélie.pdf
Files produced by the author(s)


  • HAL Id : tel-02440897, version 1


Amélie Barozet. Development and assessment of protein loop modeling methods: Application to CDR loops in antibodies. Automatic. Institut national des sciences appliquées de Toulouse, 2019. English. ⟨tel-02440897⟩



Record views


Files downloads