An atomic radii set and Generalized Born implicit solvation model trained using explicit water solvation free energy data

Abstract

Compared with the other common implicit and explicit water models, Generalized Born (GB) models can provide a fast approximation of solvation free energy that is reasonably accurate but fast enough to use in molecular dynamics (MD) simulations. This enhances conformational sampling of the solute molecules, and also can be a valuable component of multi-scale simulations. We previously developed the GB-Neck2 model, which exhibited improved secondary structure balance and was used to successfully fold a series of small proteins. More recent simulations using GB-Neck2 with updated protein models suggest that α-helices remain somewhat over-stabilized. Here, we develop a more self-consistent model, retraining both the intrinsic solvation radii as well as the GB model parameters, using the solvation free energies of an explicit water model as training references. The new radii set, named MIRO, when used with the GBNSR6 implicit solvent model leads to improved reproduction of solvation free energies calculated in explicit water. The new GB-Neck3 model shows a good balance of secondary structures: the stability of β-sheets is improved, while the previously over-stabilized α-helices became less favorable, as expected. GB-Neck3 and MIRO radii should extend the range of problems accessible to biomolecular simulation.