Paper Citing NAMD - Abstract
Juneja, Alok; Numata, Jorge; Nilsson, Lennart; Knapp, Ernst Walter
Merging Implicit with Explicit Solvent Simulations: Polyethylene Glycol
JOURNAL OF CHEMICAL THEORY AND COMPUTATION, 6:1871-1883, JUN 2010
We constructed an accurate polyether force field for implicit solvent (IS) molecular dynamics (MD) simulations that matches local and global conformations of 1,2-dimethoxy-ethane (DME) and polyethylene glycol (PEG), respectively. To make appropriate force field adjustments for IS models of PEG, we used long-term MD simulation data of 1 mu s in explicit solvent (ES) based on the most recent CHARMM35 ether force field that includes adjustments for PEG in explicit water. In IS models, competition of attractive van der Weals (vdW) interactions between solute-solute and solute-solvent atom pairs is often not considered explicitly. As a consequence, the attractive vdW interactions between solute atom pairs that remain in IS models explicitly can yield equilibrium structures that are too compact. This behavior was observed in the present study comparing MD simulation data of the DME and PEG ES model with corresponding IS models that use generalized Born (GB) electrostatics combined with positive surface energy terms favoring compact structures. To regain balance of attractive vdW interactions for IS models, we considered the IS generalized Born with simple switching (GBSW) model in detail, where we turned off surface energy terms and reduced attractive vdW interactions to 90%, or we used alternatively even slightly negative surface energies. However, to obtain quantitatively the same local and global distributions of PEG conformers as in ES, we needed additional force field adjustments involving torsion potentials and 1-4 and 1-5 atom pair Coulomb interactions. This CHARMM ether force field, specifically optimized for IS simulation conditions, is equally valid for dimeric and polymeric ethylene glycol. To explore the conformational space of PEG with MD simulations, an IS GBSW model requires 2 orders of magnitude less CPU time than the corresponding ES model. About a factor of 5 of this gain in efficiency is due to the lack of solvent viscosity in IS models.
