Re: ABF:reversibility and number of samples

From: Eric Cyr (
Date: Fri May 23 2008 - 15:11:29 CDT

I've long wondered how to verify that the PMF computed by ABF
or other sampling methods was correct. After talking to some of
the modelers who are using ABF they said they have received
comments from reviewers about how could they "know" their
profile was correct. The best thing I could think of was to run
a simulation biased with the computed profile and see if it
gives uniform sampling. When I tried this for a dihedral angle of
alanine-dipeptide for 1ns of post simulation I found my sampling
was not as uniform as I would have liked. In fact Kolgomorov-Smirnov
goodness of fit test said it was not likely to be uniform...the KS
statistic was way too sensitive. Does anyone know of another
statistic that might not be as sensitive but would still give some
indication as to whether the sampling was uniform.


On Fri, May 23, 2008 at 12:15 PM, Jerome Henin
<> wrote:
> Luca,
> I am not sure how non-uniform your sampling is. Even in the
> best-behaved systems, you never get a strictly flat histogram. What I
> usually plot is the "free energy transform", - kT * ln(sampling), and
> I compare it to the measured PMF.
> Now to add to what Chris wrote, even though the coordinate you are
> watching seems to evolves reversibly, there might be other degrees of
> freedom in its environment (maybe neighboring residues) that are
> either drifting or getting trapped. At this point the problem is
> mostly qualitatively deciding if the system is actually undergoing the
> transformation you wanted to observe (I mean in a global sense,
> besides the particular dihedral used as a RC).
> Best,
> Jerome
> On Fri, May 23, 2008 at 8:33 AM, Luca Bellucci <> wrote:
>> Dear all,
>> I used ABF to understand some aspects of
>> lingand binding mode.
>> A this moment I have a "good" energy profile but
>> I do not have an uniform number of samples:I explore along a
>> dihedral angle in the range 0-180 and I get to peaks at 20 and about 160 degrees.
>> There is a reversibility but sampling is not uniform.
>> In particular I have
>> a minimun energy that coincide with a minimun sampling.(at 60 degrees)
>> I ran my simulation (enzyme+ligand) for about 30ns and
>> I can not hope/think that things can change even if increasing
>> the simulation length of some tens of ns.
>> At this point I think about two possibilities:
>> 1-Interpretation of these results. Do you guess they can be reliable? How can I use them?
>> 2-Split the coordinate to try to reach sampling uniformity within each smaller range.
>> For instance: 0-60, 60-120, 120-180.
>> Do you have any suggestions ?
>> Luca

4330 Siebel Center
201 N. Goodwin Ave.
Urbana, IL 61801

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:47:50 CST