[gmx-users] PMF calculation by umbrella sampling simulations

Justin Lemkul jalemkul at vt.edu
Thu Apr 30 23:20:37 CEST 2015

On 4/30/15 4:12 PM, MPI wrote:
> Hi Justin,
>     Thanks for your explanation.
> (1)
> In the tutorial, then what is a good approximate value of free energy
> ?  if  one follows the protocol.   Is -37 or so kcal mol^-1  ?

That is what you should get from the tutorial.

> Here, 26 windows of COM distances are used, each for 10 ns simulation.
> (2)
> In your original paper,  -50.5 kcal mol^-1 is the free energy for
> dissociation of a single peptide from protofibril.
> In this full preparation,  you used 31-window setup, each for 10 ns simulation.
> I've been aware of the difference between 26 and 31 windows but  questions are
> First,  if one obtains a reasonable histogram like
> https://docs.google.com/file/d/0B9uTAq1wCc1fVkRPY2FrTExFcmM/edit
> ,how can one  tell whether if  the system needs more windows ?
> because these overlap windows look appropriate.

Simply looking at histograms only tells you if you're in the ballpark.  g_wham 
has a very nice set of options for doing error analysis.  You'll note there is 
minimal overlap in a few of your windows; these would correspond to regions of 
the reaction coordinate with large error.  If you do the bootstrap analysis you 
will see this.

> Second,   after full preparation, how did you know the free energy of
> -50.5 or so  is a  GOOD value ?

I did error analysis :)  See details in the paper.

> Third,  comparing  a coarse ( the tutorial) with a full preparation (
> your original paper),  it is somehow unclear that the number of
> windows can cause  a significant difference of 13 kcal mol^-1 since
> it runs independently for 10 ns in each window.
> Please see  PMF curve attached in a simulation of  26 windows, each for 10 ns.
> https://drive.google.com/open?id=0B9uTAq1wCc1fSUtwSVh4cnFpZVE&authuser=0
> The PMF curve looks reasonable except  when it converges to 37.5 kcal
> mol^-1 or so.

There is no clear energy minimum there and you haven't done error analysis, so 
"looks good" doesn't really mean anything.



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


More information about the gromacs.org_gmx-users mailing list