A fully atomistic modelling of many biophysical and biochemical processes at biologically relevant length- and time scales is beyond our reach with current computational resources, and one approach to overcome this difficulty is the use of multiscale simulation techniques. In such simulations, when system properties necessitate a boundary between resolutions that falls within the solvent region, one can use an approach such as the Adaptive Resolution Scheme (AdResS), in which solvent particles change their resolution on the fly during the simulation. Here, we apply the existing AdResS methodology to biomolecular systems, simulating a fully atomistic protein with an atomistic hydration shell, solvated in a coarse-grained particle reservoir and heat bath. Using as a test case an aqueous solution of the regulatory protein ubiquitin, we first confirm the validity of the AdResS approach for such systems, via an examination of protein and solvent structural and dynamical properties. We then demonstrate how, in addition to providing a computational speedup, such a multiscale AdResS approach can yield otherwise inaccessible physical insights into biomolecular function. We use our methodology to show that protein structure and dynamics can still be correctly modelled using only a few shells of atomistic water molecules. We also discuss aspects of the AdResS methodology peculiar to biomolecular simulations.
REFERENCES
Using the Langevin thermostat in the hybrid and coarse-grained regions, we found it necessary to thermostat at 299 K in order to have a temperature in the atomistic region of 300 K. This is due to the fact that particles which gain excess heat in the hybrid region for the reasons outlined above may not have time to become fully thermalised before they cross into the non-thermostated atomistic region. This artefact does not occur with the use of, e.g., a velocity-rescaling thermostat in the hybrid region. However, thermostats which employ rescaling require the calculation of a global temperature, or at least a temperature over some reasonably large area, which is not straightforward in the AdResS setup. In addition, such thermostats do not sample the canonical ensemble.