Exploring the Effects of Implicit Solvent Models and Parameter Variations in Molecular Dynamics via Hybrid Monte Carlo

The 3D reference interaction site model (3D-RISM) is an implicit solvent model that estimates solvation density distributions and thermodynamics in close agreement with results obtained from explicit solvent models. However, the computational demands of 3D-RISM make its use in molecular dynamics (MD) simulations, where solvation forces need to be updated at every time step, impractical. To overcome this challenge, we adopted the hybrid Monte Carlo (HMC) method to couple 3D-RISM with the fast generalized Born (GB) implicit solvent method. In our implementation, HMC generates global trial moves through MD simulations with GB, which are accepted or rejected using the Metropolis criteria for the Hamiltonian with 3D-RISM. To implement HMC, we developed a Python script that utilizes the Amber molecular modeling suite for executing MD simulations and conducting energy evaluations. To assess the effect of different GB implementations, we conducted HMC simulations of two distinct GB models (GBOBC, and GBn) with 10,000 Monte Carlo steps of 10 ns MD runs. Kolmogorov–Smirnov tests of the resulting Ramachandran plots yields P-values of >0.94, confirming that the distributions are from the same potential, showing that we are sampling from the 3D-RISM Hamiltonian, independent of the GB model. The various GB models for MD in the HMC approach had acceptance rates of 0.45-0.61, giving an effective sampling of up to 600 ns for the longest MD trajectory length. This improved computational sampling denotes a three order of magnitude speed increase for sampling over 3D-RISM with standard MD. Notably, no decrease in acceptance rates was observed when using longer MD runs, suggesting larger speedups in 3D-RISM sampling are possible.