Re: [AMBER] Ambiguity in PMF calculated from umbrella sampling of a dihedral angle

From: Jason Swails <>
Date: Tue, 1 Oct 2013 07:50:01 -0400

On Tue, Oct 1, 2013 at 7:11 AM, Mohan Pradhan <>wrote:

> Dear amber users,
> I am following the amber tutorial-
> for the
> umbrella sampling of a dihedral angle of an alpha helical peptide from cis
> to trans conformation.
> The commands and the WHAM meta file are exactly same as given in this
> tutorial. The only change is that, in the 1st simulation, dihedral angle of
> starting structure is 0 degree (from crystal structure) and then we sampled
> 3 degree windows from 0 to 180 serially; while in 2nd simulation, dihedral
> angle of starting structure is the 180 degree (structure extracted from 1st
> simulation) and then we sampled 3 degree windows from 180 to 0 serially.
> Each window is sampled for 500 ps in both the simulations. The PMF derived
> from these calculations is quite contradictory to each other (plots
> attached). Also, I ensure sufficient overlap of sampled dihedral angles
> between consecutive windows.
> Any suggestions why the PMF calculated from these 2 simulations is not
> similar (even qualitatively)?

They look qualitatively very similar. In fact, one looks like the mirror
image of the other (reflected in the Y-axis). This suggests to me that
what you are calling 180 degrees in the first plot you are calling 0
degrees in the second (and vice-versa). Assuming you got one of the X-axes
reversed, the profiles are not only qualitatively similar but probably also
agree quantitatively within the uncertainty of each window (MBAR can help
you compute uncertainties, as can bootstrapping with WHAM) -- especially
with only 500 ps of simulation per window.

Note that umbrella sampling does not have a 'directional' property to it
like other methods like steered MD. Each window is sampling from an
equilibrium distribution defined by a potential biased with an umbrella
potential along a reaction coordinate. As a result, all snapshots that you
use in the analysis should be from that equilibrium distribution (which
means you should equilibrate each window and discard these 'equilibration'


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Tue Oct 01 2013 - 05:00:13 PDT
Custom Search