Missions

The study of small chemical systems has been revolutionized in recent years by machine learned interatomic potentials (MLIP), and questions that were impossible to tackle five years ago can now be addressed. The state-of-the-art approach for MLIP applied to chemical reactivity implies performing quantum calculations on the full simulation box, which has two issues: (i) the MLIP are system-specific and cannot be easily extrapolated to a related system, (ii) it is computationally prohibitive for enzymes. Thus, enzymatic reactions are still mainly studied with QM/MM calculations or QM/MM-MD simulations. These approaches cannot be used on a large scale due to their cost, but also, they can't capture global motions.

Activités

The goal of the current project is to develop MLIP that will replace QM/MM, leading to ML/MM simulations that will be usable for enzymatic reactions. A key step in using MLIP for reactivity is having a training set that includes structures of reactant, product and transition state: this construction can be achieved through an active learning procedure such as the one included in the ArcaNN1 software that was developed in our laboratory. This project will merge an architecture that focused on ML/MM for spectroscopy2 to ArcaNN, aiming at smoothing the creation of a diverse training set with enhanced and biased sampling simulations.
As a proof-of-concept, we will study the carboxylation step catalyzed by rubisco,3 which happens to be the most abundant protein on Earth. Our goal will be to compute the free energy barrier for the wild-type and for mutants obtained through directed evolution,4 aiming at providing chemical insights in these recent results. We will then expand the study towards computing the full free energy profiles of both the carboxylation and the oxidation reactions. Finally, we will compare orthologs and explain why different species can display very different kinetic traits for rubisco.

Compétences

The candidate should have a PhD in computational (bio)chemistry and expertise in MD simulations, quantum chemistry or machine learning. Knowledge of biosystems, analysis skills (Python), scripting skills (bash and/or Python) and machine learning are assets.

Contexte de travail

The project will take place in the CPCV laboratory (ENS-PSL, SU, CNRS), hosted by the Ecole Normale Supérieure in the Latin Quarter of Paris. The theoretical chemistry group of CPCV is multidisciplinary in terms of methods and applications, and is widely recognized for its activities. The group welcomes 15 to 20 interns, PhD students, and post-doctoral fellows. During the project, we will collaborate with Guillaume Stirnemann and Damien Laage who will bring their expertise in ML, and with Julien Henri who will bring his expertise on photosynthesis.