Hi Michael 
> I assume that the value you use (0.2394) is a consequence of
> the algorithm and not optimized for any particular kind
> of system, or is it ?
It's been optimized using nonpolar solvation forces for protein
structures but found to give accurate nonpolar solvation forces for
RNA as well.
>> We've found the optimal parameters to work well in other force
>> calculations as well.
>
> I am primarily interested in energies, rather than forces,
> The latter being derivaties of the energies, means that, if you
> optimize your model to obtain correct forces, the resulting
> energies might still be off by some constant value.
Right. We're working on applying the approaches outlined in
Swanson JMJ, Wagoner JA, Baker NA, McCammon JA. Optimizing the
Poisson dielectric boundary with explicit solvent forces and
energies: lessons learned With atomcentered dielectric functions. J
Chem Theory Comput, 3, 170183, 2007. (http://dx.doi.org/10.1021/
ct600216k)
to nonpolar forces to address this issue.
> Can I expect the nonpolar energies resulting from your
> algorithms to be consistent with the electrostatic energies
> I get from PB calculations (provided of course the
> PB parameters, radii, charges, epsilon_P are appropriate)
No, not necessarily. We mainly released these algorithms based on
user requests  not because we have complete sets of parameters for
them. Any new systems will definitely require testing and possibly
reparameterization.
> Talking about PB parameters, I've done some work
> comparing binding energies from PB calculations combined with
> simple DeltaSASA based nonpolar energies, Notwithstanding the
> known shortcomings of the latter model, did you find any particular
> set of PB parameters  e.g. Bondi vs Parse vs some, possibly
> scaled, force field based radii, or AM1BCC vs force field based
> charges, etc  to perform better than others ?
No  based on our force analysis, all SASA methods performed very
badly for a variety of radii.
> In my, but small, test set CHARMm based radii plus CHARMm (mind
> the lower case m) peptide charges combined with ESP charges for
> ligand and epsilon_P=1 seemed to perform best, closely followed
> by the parameters suggested by Nina and Roux, 1997. However, both
> these two, and a number of other parameter sets/combinations I
> tried did NOT provide really satisfactory results in terms of
> reproducing experimental binding affinities.
> Again, this might be due to the poor model for the nonpolar
> terms. Considering this it is, of course, very important
> to improve such nonpolar models, as you did in your recent
> publication  and I wonder if, from this work, you gained any
> insights into not only nonpolar forces, but also in the
> appropriate parameterization of PB models ?
The paper I cited above might have what you're looking for. Note
that one of the conclusions of the paper was that splinebased
dielectric definitions have some potential caveats  this is
important to keep in mind when comparing to more detailed models!
Thanks,
Nathan

Associate Professor, Dept. of Biochemistry and Molecular Biophysics
Center for Computational Biology, Washington University in St. Louis
Web: http://cholla.wustl.edu/
