Développement d'approches prédictives pour la cinétique d'association biomoléculaire
Karen Palacio Rodriguez · HAL (Le Centre pour la Communication Scientifique Directe) · 2022
Atomistic computer simulations of rare events have three paramount goals: predicting detailed mechanisms, free energy landscapes, and kinetic rates of transformation processes like phase transitions, chemical reactions, biomolecular folding, or association. In real-life applications, all of these tasks are cumbersome and require intensive human and computer effort, especially the calculation of rates. The difficulty resides in the gap between the long time scale associated with such processes, also known as rare events, and the short time scale that is accessible by molecular dynamics simulations. Enhanced sampling techniques can accelerate the exploration of high-free energy regions, adding external forces to the system to pull it out of free energy basins or focusing sampling on the transition region and efficiently exploring transition paths. These techniques allow to reconstruct mechanisms and free energy landscapes for a wide range of activated processes in physics, chemistry, and biology. Methods aimed at accurate kinetic rates are at present less mature and still require large computational effort and/or rely on ideal collective variables. We developed two efficient methodologies for the prediction of transition rates from molecular dynamics simulations in combination with enhanced sampling techniques. Both strategies only require sets of short simulations, which allows exploiting the parallel capabilities of current supercomputers. On one side, we use metadynamics, a widely used enhanced sampling technique that adds a time-dependent bias potential to the system, disrupting its dynamics. We overcome this limitation by developing a method based on Kramers’ theory for calculating the barrier-crossing rate when a time-dependent bias is added to the system. We tested this method in a benchmark system and apply it to complex all-atoms simulations, showing that we are able to extract the rate and measure at the same time the quality of the collective variables for processes where Kramers' theory holds. On the other side, transition path sampling trajectories are the golden standard to access mechanistic information: we demonstrate that they also encode accurate thermodynamic and kinetic information, that can be extracted by training a data-driven overdamped Langevin model of the dynamics projected on a collective variable. We also tested this method over benchmark systems to establish validation criteria for the accurate time resolutions that yield markovian behavior and apply it to complex all-atoms simulations to recover free energies, position-dependent diffusion coefficients, and rates. Overall, these new theoretical tools that can be freely downloaded from GitHub make efficient use of computing resources providing simple procedures to accurately predict kinetic rates, and could be suitable for applications far beyond the field of biomolecular association.