[gmx-users] dispersion correction and heterogeneous systems

David Mobley dmobley at gmail.com
Wed Jan 31 03:09:11 CET 2007


> Does anybody have any experience with DispCorr=EnerPres and a
> heterogeneous system (e.g. aggregation of detergents to form micelles)?
> What about membranes? Even if it is not a perfect method, is an
> imperfect assumption of homogeneous LJ density beyond the cutoff better
> than entirely ignoring it?

Michael Shirts and I are working on dispersion corrections for
heterogeneous systems in the context of ligand binding. Actually,
we're working on a paper, of which we'll probably have a preprint in a
couple weeks which I can send if you are interested. The upshot here
is something like this: DispCorr=EnerPres tends to help in both
proteins and water, because otherwise contributions to LJ interactions
from outside the cutoff are totally neglected.  In the context of
ligand binding, though, it turns out to be necessary to go somewhat
further than this (at least for absolute binding free energies) --
because the uniform density assumption would give a comparable
correction for the systems where the ligand is alone in solvent,
versus in the protein. In reality, when the ligand is in the protein,
it is surrounded by a higher density of LJ sites, and should feel
greater attractive forces even from stuff outside the cutoff. The way
we end up handling this is using a reweighting scheme where we save
trajectory snapshots run with normal cutoffs, and then reprocess them
using longer cutoffs and reweight observables computed from the
trajectories with the short cutoffs to properly account for these sort
of differences.

> Currently I am using NPT with DispCorr=No. I am working my way toward
> free energy calculations in which I gather that I must use
> DispCorr=EnerPres. However, for simulations that are not intended for
> free energy calculation is DispCorr=No quite acceptable?

I think it probably depends on what you're trying to do. It's probably
worth noting that DispCorr actually affects pressures as well, so you
might need it to get densities right.

> Papers that apply long range corrections as the best method:
> Shirts and Pande, JCP 122, 134508 (2005)

You might also want to look at Michael Shirts' dissertation, the
chapter on FKBP. You can also contact me or him off-list for more
details on what we've finally decided on.

I don't know most of the other papers you mention -- will have to have
a look. I do know this one:

> Horn, Swope, and Pitera, JCP, 120, 9665-78 (2004) use it in their TIP4P-Ew

As I said, to some degree, I think it depends on what you're trying to
compute -- but in my opinion at the very least you should be using
this analytical dispersion correction that assumes uniform sites --
it's better than NO dispersion correction. If you go the direction of
ligand binding you may want to do something more sophisticated we can
talk about off list.


More information about the gromacs.org_gmx-users mailing list