In implicit solvent models, the non-polar component of solvation is often assumed to correlate with the surface area and/or the volume based on theoretical arguments relating to cavity creation cost22 –26 (link). To explore this we computed the solvent accessible surface area and volume for all of the solutes considered here using GROMACS tool
g_sas with a probe radius of 1.4 nm.
We also further dissected the non-polar part (due the Lennard-Jones interactions) into repulsive and attractive components using the Weeks-Chandler-Andersen (WCA) separation27 . To do this, we implemented the WCA separation in a modified version of GROMACS 3.3.1.45
In our main study, we simply computed the total non-polar component and retained the trajectories. The attractive component for each solute was then obtained by applying the WCA separation to stored trajectories of the fully interacting solute, and reprocessing these simulations with the attractive interactions turned off to re-evaluate the energies. We computed the free energy for turning off the attractive interactions using exponential averaging (the Zwanzig relation28 ) and standard error analysis. This assumes that phase-space overlap is good between the ensemble where the solute has attractive interactions with water, and that where it does not. Error analysis should tell us if this is not the case. We further tested this by re-computing the attractive contribution using simulations at series of separate λ values (where λ modifies only the attractive interactions) for selected solutes (phenol, p-xylene, pyridine, and toluene) and found that computed free energies were within uncertainty of the values computed using exponential averaging, indicating overlap was sufficient.
With these attractive components, we then obtained repulsive components by subtracting the attractive component from the total non-polar component. This probably results in slightly larger uncertainties in computed repulsive components than would have resulted from computing the repulsive component separately, but it also saves a large amount of computer time since we had already computed the total non-polar component, and the repulsive portion of the calculation is the most difficult to converge.
We also further dissected the non-polar part (due the Lennard-Jones interactions) into repulsive and attractive components using the Weeks-Chandler-Andersen (WCA) separation27 . To do this, we implemented the WCA separation in a modified version of GROMACS 3.3.1.45
In our main study, we simply computed the total non-polar component and retained the trajectories. The attractive component for each solute was then obtained by applying the WCA separation to stored trajectories of the fully interacting solute, and reprocessing these simulations with the attractive interactions turned off to re-evaluate the energies. We computed the free energy for turning off the attractive interactions using exponential averaging (the Zwanzig relation28 ) and standard error analysis. This assumes that phase-space overlap is good between the ensemble where the solute has attractive interactions with water, and that where it does not. Error analysis should tell us if this is not the case. We further tested this by re-computing the attractive contribution using simulations at series of separate λ values (where λ modifies only the attractive interactions) for selected solutes (phenol, p-xylene, pyridine, and toluene) and found that computed free energies were within uncertainty of the values computed using exponential averaging, indicating overlap was sufficient.
With these attractive components, we then obtained repulsive components by subtracting the attractive component from the total non-polar component. This probably results in slightly larger uncertainties in computed repulsive components than would have resulted from computing the repulsive component separately, but it also saves a large amount of computer time since we had already computed the total non-polar component, and the repulsive portion of the calculation is the most difficult to converge.