From: Jérôme Hénin (jerome.henin_at_uhp-nancy.fr)
Date: Mon Jan 10 2005 - 15:57:01 CST
You are pointing out an issue I am concerned about too, and possibly the main
weakness of the current alchemical FEP module. I mentioned it briefly to
people in the Schulten group when I last visited them, but since it was not
my main purpose, we did not come to any useful conclusion on that.
As you say, implementing a separation-shifted potential does not raise any
conceptual problem. A programmer familiar with NAMD and in particular the
force-field implmentation should only need a reasonable amount of work to
get it done. The problem boils down to the following: the group of people
in the community that actually master most of the NAMD code is, as far as I
know, small. Possibly very small.
When alchemical FEP was contributed by our group, the post-doc who coded it
had to spend a long time just to get familiar with the NAMD code. Now we
would very much like to add such improvements as soft-core potentials, but we
do not currently have the manpower for it. If you can contribute to an
implementation, it will definitely be a significant advance for all users of
this method. The best way I can imagine for now is to involve Klaus
Schulten, who has access to both manpower and, obviously, technical knowledge
Anyway, we commonly use that code, and there are ways to try to circumvent the
end-point problem. The main point is a careful design of the windows,
employing a large number of smaller and smaller windows as lambda approaches
1. I can send you an example config file tomorrow.
Another improvement I have thought of recently is to replace the FEP formula
with an overlap sampling method, for instance Bennett's method (see Lu, Singh
and Kofke, J Chem Phys 118 (2003)). This significantly improves convergence
for FEP calculations with a limited sampling.
This is doable with the current version of the code, by post-processing the
energy data output by the FEP module. I have a small program that does it.
Anyone interested can just email me.
Unfortunately, it is of no help with regard to the end-point catastrophe,
because overlap sampling is often not feasible in the very last window. I
will not expand on the reasons here - they are probably obvious to
specialists and not so interesting to non-specialists.
Thanks for raising that point!
Le Lundi 10 Janvier 2005 18:12, Philip Fowler a écrit :
> Hi everyone,
> I have already run one calculation of a difference in binding free energy
> using the FEP module within NAMD. I am about to start calculating a second
> and need your help .
> In my first calculation (and this will happen again) I encountered the
> well-known "end-point catastrophe" where, because the van der Waals
> component is scaled linearly by lambda, the repulsive term remains and
> prevents the water exploring the void created when the atom has truly
> One solution to this is to follow the "separation-shifted" method of
> Zacharias, Straatsma & McCammon (J Chem Phys 100 9025 (1994)) where, for
> nonbonded interactions with alchemicalk atoms, the conventional vdW LJ form
> V = (1-lambda) [ A/r^12 - B/r^6 ]
> is replaced by
> V = (1-lambda) [ A/(r^2 +dlambda)^6 - B/(r^2 +dlambda)^3 ]
> (where dlambda is a supplied parameter)
> which is the same when lambda = 1. This form allows a much smoother
> transition as lambda=0,1 is approached.
> Now the questions are,
> 1) how and where would one change this in the source code? I have had a
> rummage but understanding the object structure would take me more time than
> I would like (I am not an expert at C++). In principal the logic is simple:
> for alchemical nonbonded vdw interactions, apply the new functional form,
> for everything else, keep it as it is.
> 2) would this be useful to anyone else?
-- Jérôme Hénin Equipe de Dynamique des Assemblages Membranaires Université Henri Poincaré / CNRS UMR 7565 B.P. 239 54506 Vandoeuvre-lès-Nancy Cedex Tel : (33) 3 83 68 43 95 Fax : (33) 3 83 68 43 71 http://www.edam.uhp-nancy.fr/
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:39:06 CST