Re: FEP tutorial graph question

From: Jerome Henin (
Date: Wed Apr 29 2009 - 18:36:35 CDT

Hi Chris,

Chris' question about the version of NAMD is important. I'll try to
address the other points.

Moving the alchemical parameter backwards or inverting the FEP flags
gives the same result - it's just a matter of convention. Internally,
each group has a coupling parameter that is either lambda if the group
is appearing or (1 - lambda) if it is vanishing.

Combining data from the forward and reverse transformations should be
done using the appropriate formalism, that was best described by Lu
and Kofke there:
Just averaging the free energy differences generally gives very poor
convergence. In contrast, Bennett's estimator, although somewhat more
involved, has been shown to be optimal (in the sense of minimum
variance). Other "overlap sampling" methods converge pretty well too.
David Kofke's papers explain all that.

Restraints on the perturbed groups are mostly useful when computing
binding free energies by the double-decoupling method. For the theory,
see Gilson et al:
and this paper by Deng and Roux:
(as well as a number of related papers by the McCammon and Roux groups).


On Wed, Apr 29, 2009 at 4:35 PM, Christopher Hartshorn
<> wrote:
> Hello all.  I am having problems figuring out why my plot of dG vs. lambda
> looks nothing like the plot in the FEP tutorial writeup.  I understand that
> I will not necessarily get the same ddG because of various issues, but what
> I get is nothing like anything I have seen (including all published
> sources).  I have followed the tutorial to the letter (minimize,
> equilibrate, and FEP with 100K iterations/lambda step) with the tyr2ala
> part, but when I plot up the results they look completely different (as in
> they curve up then down then up and then down again not the smooth curve I
> see in most).  I have a small jpeg (73KB) of the graph I can send if need
> be.  I generate these plots by using the .fepout files lines that contain:
> #Free energy change for lambda window [ 0 1e-05 ] is 0.0253223 ; net change
> until now is 0.0253223
> And then simply plotting the step versus the net change until now (e.g.
> 1e-05 vs. 0.0253223 in the above example).  Am I doing this wrong?  Plotting
> up the wrong data?  I began doing this because I was seeing the same results
> when I plotted up some FEP calculations that I am doing on systems of mine
> and wanted to see if I could even get the tutorial done correctly (which I
> clearly can not, meaning I must be missing some point here because none of
> my graphs of dG vs lambda have the smooth curves I have seen every where
> else).
> Also, as a side note I have read of people who do an averaging of both
> directions to get better results.  But I am not sure if they mean forward
> direction= (going from 0 to 1) and then 1 to 0 for reverse direction or if
> they mean that forward=fading out (beta column=-1) interaction of molecule
> of interest from 0 to 1 and reverse=fading in (beta column=1) molecule of
> interest from 0 to 1 (although, I guess those are probably the same things
> my question is with respect to NAMD and which is the better way, if any, in
> practice).  Is this averaging actually better then going in the "more"
> correct direction to begin with (e.g. the direction of lower entropy)?
> Finally, and I will stop here, I thought I had read of people putting
> restraints on (or even fixing) the molecule that is being faded in/out for
> the FEP calculation.  Is this normal practice?
> Thank you very much in advance.
> Chris Hartshorn

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:41 CST