Quantitative Molecular Simulation

Here I propose to build a new force field (FF) in a systematic manner where parameters are derived through Bayesian optimization and machine learning. Derivation of the FF will be done in a step-wise fashion.

Polarizability and electrostatic parameters have been done already by developing and implementing algorithms to determine partial charges that reproduce the electrostatic potential around compounds. Here I will focus on the intramolecular potential and force constants and parameters for Van der Waals interactions.

FF methods are subject to stochastic and systematic errors in their predictions as well as errors due to uncertainties in the parameters and here we will determine the impact of uncertainties in parameters on predictions.

Finally, we will implement a method for stochastic proton hopping into GROMACS based on the new force field which will enhance the applicability of simulations greatly.

PI will train a Ph.D. student and one postdoc or researcher in model development through machine learning, critical evaluation of simulation results, software development and high-throughput application of simulations.

The project will increase the predictive power for applications of FF methods but also the robustness by giving realistic error estimates to predictions. Implementation in GROMACS will speed-up adoption of the models.

Dissemination through our website at http://virtualchemistry.org will help outsiders to evaluate the state-of-the-art in FF methods as well.