[AMBER] Adaptively Biased Molecular Dynamics

From: Aron Broom <broomsday.gmail.com>
Date: Tue, 6 Dec 2011 18:39:31 -0500

Hello AMBER users,

I've been performing free energy calculations using MetaDynamics in NAMD
and thought that I would try AMBER and its related Adaptively Biased
Molecular Dynamics (abmd) module. I looked at the user guide, tutorial
#12, and also at the referenced paper where this method was developed. I
am applying the method to the extremely simple test case of a water dimer
in a vacuum, attempting to generate the PMF for the reaction coordinate of
the distance between the two waters. I've run into three snags with the

1) From the documentation provided it seems like you can only define
reaction coordinates between single atoms rather than centers of mass. For
instance, for the DISTANCE collective variable, the definition says "type =
DISTANCE : distance (in ) between two atoms whose indexes are read from
the list i.", and the corresponding example also shows this same limitation:

 type = DISTANCE
 i = (5, 9)
end variable

It is generally accepted that defining reaction coordinates between single
atoms is a very bad practice in many cases, and that reasonable accuracy is
only obtained when comparing the center of mass of many atoms. I realize
that for the case I mention here of a water dimer its mostly irrelevant,
but this is just a test case and I'd like to eventually compare against a
protein with small molecule ligand in which case it is quite relevant. Is
there a way to use the center of mass rather than an individual atom?

2) I ran the water dimer test just using the oxygens for each water and the
PMF is semi-reasonable, but the entire simulation suffers from the fact
that the distance can become larger than the range I'm interested in. For
instance for the abmd I have the following reaction coordinate definition
in my *.mdin file:

 mode = FLOODING

 monitor_file = 'abmd.txt'
 monitor_freq = 100

 timescale = 10.0 ! in ps

  type = DISTANCE
  i = (1, 5)
  min = 0.0
  max = 10.0
  resolution = 0.05
 end variable
end ncsu_abmd

So as you can see I don't care once the distance gets past 10 angstroms,
and I'm doing the whole thing in a vacuum without periodic conditions, so
in fact I don't want it just flying off into space, but instead need it to
stay within that range and continue sampling until I get a nice smooth
PMF. In the case of tutorial #12 this isn't an issue because it studies a
dihedral angle which is naturally bound from -180 to +180 degrees. When I
run my simulation I get about 50ps worth of information which starts to
build up a rough PMF, and then the distance gets past 10 angstroms at which
point the effect of building up the repulsive gaussians as part of the abmd
routine means that it is pretty much never coming back. In NAMD the
metadynamics module contains a simple variable which sets a hard lower and
upper potential wall for the reaction coordinate. I can't seem to find the
equivalent in AMBER. Using a harmonic restraint isn't a reasonable option
because that would bias the whole region and then I'd have to do something
more complicated to unbias it later which seems like a needless pain. Is
there a way to apply such a hard wall?

3) The references article describing the abmd procedure indicates that
there are only two variables that control the simulation, and one of them
is the flooding time scale. Nowhere in that article can I find a useful
definition of that variable. Is this the delay between applying a new
repulsive gaussian? Should this be set to a small fraction of the total
simulation time?

Thank you for your time,

AMBER mailing list
Received on Tue Dec 06 2011 - 16:00:02 PST
Custom Search