Colloquium Speaker:

Prof. Marie-Pierre Gaigeot Université Paris-Saclay, Univ Evry
CNRS, LAMBE UMR8587
Evry, France

Topological molecular graphs (2D-MolGraphs) for post-processing molecular dynamics simulations

Abstract:

Algorithmic graph theory, a branch of AI, is powerful for the analysis of molecular simulations. In particular, a 2D-graph encodes topological properties of matter through vertices and edges that report on the specific (pairwise) interactions between the vertices. At the molecular level of representation, the vertices are usually associated to atoms while the edges report on the interactions between the atoms, e.g. chemical bonds and intermolecular interactions. Modelling the interactions between the atoms of a molecule by a 2D-graph is nowadays well established, examples include the classification of similar molecules according to their topology, the prediction of patterns in biological molecules, the prediction of the 3D structure of small molecules, etc. Molecular graphs are also commonly used in supervised machine learning algorithms, especially for neural network schemes. In this presentation, I will review graph theory-based methods that were recently developed in our group for interpreting and post-processing molecular dynamics (MD) trajectories generated at the atomistic level of representation. We have developed 2DMolGraphs and associated algorithms that I will show to provide a direct and fast methodology for the identification of the 3D-structures of the conformers of molecules and assemblies of molecules sampled over time in MD trajectories, with an easy identification through graphs of transitions of the interconversions between the conformations. I will present the versatility and transferability of our topological 2D-MolGraphs with applications to the conformational analysis of gas phase molecules and inhomogeneous aqueous solid interfaces from molecular dynamics trajectories. I will also show our current development of a coarse grained topological 2D graph based on H-bonded rings/cycles and their polymorphism for biological molecules. Also, applications of these graphs to chemical reactivity will be shown. An even more complex challenge is to predict 3D structures from topological 2D graphs. Our first attempts to tackle such a challenge will be presented with the development of game theory and reinforcement learning methods for predicting the 3D structure of a gas-phase peptide.