Frédéric Cazals
Centre Inria d’Université Côte d’Azur
frederic.cazals [at] inria.fr
Abstract
Protein dynamics are key to protein functions, with action modes ranging from subtle motions impacting thermodynamics, to large amplitude conformational changes involved in complex multi-body mechanisms. While the prediction of (well) folded structures may be taken as an achievement in the deep learning era with Alphafold2 and the like, predicting dynamics essentially remains an open problem.
This talk will review recent work in this realm, based on novel insights on loop closure techniques coupling kinematic models in high dimensional dihedral angle spaces, and Monte Carlo Markov Chain sampling techniques of the Hit-and-Run type. Along the way, I will discuss connexions with other problems, including high dimensional volumes and densities of states, as well as mixture models in flat tori to capture couplings between torsion angles.
These ingredients will make us ponder on the opportunity to be optimistic regarding the accurate and fast prediction of protein dynamics.
Keywords: proteins, conformational changes, loop closure, sampling, Monte Carlo Markov chains.