Re: Alchemical FEP NAMD tutorial - problems with reproducing the results

From: Jawahar Neelankatan (
Date: Wed Feb 20 2008 - 14:48:54 CST

Hi Chris,
Thank you very much for your email. I was able to use psfgen with the correct patches (ACE at N-terminal and CT3 at C-terminal) to generate the correct structure. I used this structure to do a simulation (in vacuo) using the same parameters listed in the tutorial (300K, 10 ps^-1 damping coefficient, cutoff electrostatics, rigidBonds set to 'all', and integration timestep of 0.5 fs) however I got a dG of 3.97 kcal/mole as opposed to the value of 4.4 kcal/mole reported in the tutorial.
I suspect the reason for this discrepancy might be that the initial coordinates and velocities of my system differ from that used in the tutorial. Is there a long enough simulation (longer than 2.5 ns) I can do that would more closely approximate the tutorial's results ? Also, I noticed the huge temperature fluctuations even though I used the relatively high damping coefficient of 10 ps^-1 (as instructed in the tutorial).
Thanks again for your email.


Chris Chipot <> wrote: Dear Jawahar,

> (1) Setup phase:
> The Alchemical FEP tutorial includes some setup files, e.g. a topology
> file containing the hybrid topology of tyrosine and alanine
> ( and the hybrid structure of the whole system (tyr2ala.pdb)
> The first thing I noticed about the tyr2ala.pdb file is the strange
> atoms of the first and last alanines. I'm referring to non-traditional
> atoms going by names such as 'CAY', 'CY', 'OY', 'HY1', etc. Alanine has
> only 3 carbons but the alanines in tyr2ala.pdb had 5. I also noticed
> that the topology file included in the setup files
> ( does not support this odd version of alanine,
> and, when this topology file is used with psfgen to generate a psf file
> (using the setup.pgn script given in the tutorial), psfgen basically
> just complains that it cannot recognize these non-traditional atoms, so
> it generates the usual alanine. So the alanines generated by psfgen were
> different (in the number and nature of atoms) from the alanines in
> tyr2ala.pdb. It was that psfgen-generated structure that I used to do my
> simulations, and I got very different results.

This is not an odd version of alanine. The tripeptide is blocked at its
N- and C-termini by means of an acetyl, i.e. CH3-CO (ACE) and an amine,
i.e. NH-CH3 (CT3) moiety. The latter are defined as "patch" residues in
Charmm. When you build the peptide, please note the termination default
in your topology file. This default can be overruled in PsfGen using the
"first" and "last" keywords. I suspect that your default is NTER and
CTER, i.e. charged -NH3+ and -COO- groups.

> (2) For the vacuo simulation of Ala-Tyr-Ala, I used the parameters
> listed in the tutorial and did a minimization and short heating
> initially, before even doing any FEP runs. I noticed that the heating
> step produced strange results. I used langevintemp to heat up the system
> slowly in 2K increments from 0K to 300K. The temperature was never
> stable, and fluctuated widely no matter how slow I heated it, in fact,
> after heating up to 300K and letting it equilibrate for 20 ps, I was
> still seeing temperature fluctuations as huge as 150K. I experimented
> with all sorts of heating times and heating increments and I still
> couldn't get rid of the wild temperature fluctuations so I gave up and
> decided to move to the FEP phase of the simulation. I used the .vel,
> .xsc and .coor files from the minimization and heating run as the
> starting configuration of the FEP run. I ran the FEP run exactly as
> specified in the tutorial but the result I got for the free energy
> difference for the isolated (in vacuo) state 3.6 kcal/mol, which is
> different from the value of 4.4 kcal/mol obtained in the tutorial.

Temperature drifts, especially in the gas phase, are common if use is
made of too loose a coupling to the heat bath, i.e. if the friction
term is too small.

Although I cannot comment on the starting point that you utilized, I
would like to emphasize that the computed free energy will inherently
depend upon your initial set of Cartesian coordinates and momenta. If
you run the calculations twice, starting from distinct points of phase
space, it is likely that you will obtain two different free energy
differences based on finite-length simulations, i.e. limited sampling.



Chris Chipot, Ph.D.
Equipe de dynamique des assemblages membranaires
Unité mixte de recherche CNRS/UHP No 7565
Nancy Université Phone: +33 (0)3-83-68-40-97
B.P. 239 Fax: +33 (0)3-83-68-43-87
54506 Vandœuvre-lès-Nancy Cedex


     Science without management is worse than management without science

                                                        N. G. van Kampen

Looking for last minute shopping deals? Find them fast with Yahoo! Search.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:47:34 CST