[gmx-users] g_wham PMF profile

Jochen Hub jochen at xray.bmc.uu.se
Tue Mar 15 23:27:06 CET 2011

> The reliability of the PMF curve depends on the reliability of 
> sampling.  If you're over-sampling in some regions along the reaction 
> coordinate and under-sampling in others, then the weighting is 
> probably wrong and the result inaccurate.  I can't tell exactly from 
> your description what's going on, but if you have largely overlapping 
> histograms, then your window spacing and/or force constant settings 
> are inappropriate.
I don't really agree with the last point (with the first I fully agree). 
In my experience it helps a lot if you have multiple overlapping 
histograms. If each histograms overlaps only with the two neighbors, you 
typically get large uncertainties - even worse, since you usually don't 
know the autocorrelation time, you cannot even estimate your 
uncertainty. In addition, I don't see any reason why WHAM should have 
problems in case of largely overlapping histograms.

My strong suggestion is to use as many histograms as possible, such that 
multiple histograms overlap. Make sure (if possible) that each histogram 
is independent by starting from independent initial frames (if these are 
available). Then you can use the boostrap of complete histograms (using 
the so-called Bayesian bootstrap) to compute the uncertainty:

g_wham -nBootstrap 100 *-bs-method *b-hist ...

As David just pointed out, please have a look into our recent paper: 

Jochen S. Hub, Bert de Groot and David van der Spoel: g_wham - A free 
weighted histogram analysis implementation including robust error and 
autocorrelation estimates J. Chem. Theory Comput. 6 pp. 3713-3720 (2010)




Dr. Jochen Hub
Computational and Systems Biology
Dept. of Cell&  Molecular Biology
Uppsala University. Box 596, 75124 Uppsala, Sweden.
Phone: +46-18-4715056 Fax: +46-18-511755

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20110315/35ca7531/attachment.html>

More information about the gromacs.org_gmx-users mailing list