[gmx-users] Umbrella Sampling and Histogram

Agnivo Gosai agnivogromacs14 at gmail.com
Fri Feb 20 21:48:06 CET 2015

Dear Users

I am also attaching a picture of the PMF. It looks fine to me and I would
also like to know your views on it.

If you look at the image you will find that there is a "valley" around 3.5
angstrom mark. It is because the data from the simulation for that
particular window was not added to WHAM. Also I am running a few
simulations still , to add 9 more windows towards the end of the
trajectory. You will find a dangling tail there.
My final g_wham run will consist of 43 windows having 0.1 nm spacing. Now
it is having 33 windows.

>From the above plot it seems that the PMF is flattening out after a certain
COM distance between the two molecules.

For error analysis I did a Bayesian bootstrap analysis with -nBootstrap 50
-bins 50, with a default tolerance of 10^-6. I am attaching the bsProf and
bsRes plots below:

The other simulations will take another 20 hours to complete , so I will
send another mail with the PMF , Histogram and Bootstrap data for a quick

I have read the g_wham paper ( Hub et. al. JCTC 2010) and it suggests that
Bayesian bootstrapping methods provide reasonable error estimate. The bsRes
plot shows the standard deviation as a red line. From the xvg file it seems
to be around 1.0 kcal/mol. Now in few of the papers calculating PMF using
Gromacs and g_wham I did not find anything on these analyses except the
histograms in some of them.
So how should I interpret this error ?
Based on the present figures can anyone comment on the convergence and
error ??

On 2/19/15 9:50 PM, Agnivo Gosai wrote:

> Dear Users
> This is my earlier post wherein I received useful suggestion from Dr.
> Lemkul.
> http://comments.gmane.org/gmane.science.biology.gromacs.user/74659
> Earlier for my system , I was doing a 1 ns Umbrella Sampling run for each
> window and my histogram looked like this :- ​
> <https://docs.google.com/file/d/0B-U8uULVZjfRQXcyUkI2bXhBcVE/
> edit?usp=drive_web>
>> Now, I , kept all other stuff ( e.g. spring constant same ) only increased
> my simulation time to 10 ns. My histogram now looks like this :-
>>   hist
> <https://docs.google.com/file/d/0B-U8uULVZjfRcVhLN196TXdxcm8/
> edit?usp=drive_web>
>> I think the shape of the histograms is OK now, they dont appear rocky and
> broken. I suppose I need proper overlap at present with regards to window
> spacing and all.
That's true.

Kindly review and suggest.
How does the PMF look?  Is it converged?  What are the error estimates from
it?  Presumably you're interested in free energy differences, so that's
what you should start by examining.



Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441

Thanks & Regards
Agnivo Gosai
Grad Student, Iowa State University.

More information about the gromacs.org_gmx-users mailing list