On Wed, Jan 31, 2018, Muthukumaran Rajagopalan wrote:
>
>
> I am performing thermodynamic integration for the disappearing of few
> ions in protein complex.
> I am using Sander to convert an ion (initial state) to a dummy atom
> (final state).
This sounds like something for which you should use softcore: we've tried for
a long time to avoid using dummy atoms. Think of it this way: rather than
"disappearing" the ions, you rather want to "decouple" the ions: the ions are
still there (no dummy atoms), but in the final state they no longer interact
with their environment.
> I am modifying both the charge and vdw parameters at the same time to
> make the ion disappear.
This is dangerous without using softcore.
>
> I am following the tutorial in the below link as a starting point.
www.rosswalker.co.uk/tutorials/psc_workshop/Tutorial_seven/
Sigh...proves you can never escape your past in the modern era. I prepared
that tutorial, but note the date at the top says "2005", and the example
files were run in October of 2002. Since then we have learned that
divergences in "simple" TI when particles disappear are better handled by
using softcore than by the "klambda" approach. I think this is especially
true when trying to remove charge and vdw at the same time.
However, I don't think you have been wasting your time, since you have clearly
learned a lot about the innards of what is going on with free energy
calculations. Your questions are good ones, but I think the only good answer
is to move to an alternate approach.
...good luck....dac
[For the greybeards on the list: at the tutorial notes, the point of these
calculations was to demonstate (mainly to myself!) the extent to which Amber
could reproduce the early "extremely precise" free energy results reported by
Michael Shirts. The web page cited above is also on the ambermd.org site, but
with a "deprecated" title, and an introduction explaining that it is outdated.
But I think it is still a good educational exercise, prompting people to look
carefully at how dV/dlambda changes with lambda, and what the problems are at
endpoints -- problems that led most in the field to move towards soft-core
approaches, as discusse above.]
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Feb 01 2018 - 04:30:04 PST
