Développement de méthodes mathématiques pour l'analyse de trajectoires conformationnelles en dynamique moléculaire

Molecular dynamics (MD) simulations can produce nowadays huge amount of data using high-throughput CPU/GPU clusters. However, in addition not to consider the kinetic aspect of the data, the routine use of MD simulations for a study of real macromolecules present some drawbacks. The RMSD approach remains too vague for the description of local equilibrium, whereas classical clustering methods can provide artifacts: the both points of view share a limited exploitation of the information held by the MD data regarding the dynamics of the underlying process. We will present here a work towards the dynamics modelling of such trajectories, and we will show to what extend they can me described as resulting from an SDE, or better, a Langevin equation. Furthermore, we present a new method addressing the detection of local equilibrium, the κ-segmentation algorithm, allowing to detect and quantidy local equilibrium, while avoiding the detection of artifacts, and thusly, making up for the downsides of the quoted methods.