.jpg)
Classical molecular simulations require empirical parameters to describe the interactions between particles. In principle, a good set of parameters should be able to give the correct structural and thermodynamic properties. Determining these parameters is a challenging optimization problem, as one has many parameters to optimize, many target properties in the objective function, and ideally a final set of parameters that is transferable to similar systems.
In this work, the authors develop a differentiable molecular dynamics software (∂-HylleraasMD) and a protocol to parametrize force fields based on the Hamiltonian hybrid particle-field formalism. Using automatic differentiation as implemented in JAX, a protocol with parallel replica simulations, and many target systems is developed. The results show that the method is robust, and a set of parameters for phospholipids is optimized.
Publication: July 4, 2024; https://pubs.acs.org/doi/10.1021/acs.jcim.4c00564