Re: [AMBER] Replica exchange umbrella sampling

From: Fabrício Bracht <>
Date: Sat, 5 Oct 2013 04:29:29 -0300

Hi Jason. I see what you mean about replica umbrella. I think there is no
way around then. Since you mentioned the adaptively biased md options, let
me point a few problems I had with that in my past attempts to use it. The
whole ncsu module in amber is very interesting to read and it would seem
that it works fine. I probably does, for some cases, but when dealing with
chemical reactions, I was not able to set any simulation that actually gave
me anything useful. The file, for example, is a black box. The
ncsu-umbrella-slice utility is long gone by now and that means that both
umbrella and flooding simulations are quite difficult (if not impossible)
to perform correctly. I also found that the flooding mode does not prevent
solvent or reactants to simply flow away from the reaction center. The
restraints you impose with the "i" and "r" variables do not seem to do what
they are intended to. They did not work in any of my tests, at least. This
could also mean that I am a "very limited user" and I am doing things
completely the wrong way. But, since the nmr restraints mode of amber work
fine and I was able to get nicely shaped free energy graphics from that, my
guess is that I am not so limited after all. Judging from the amount of
comments on ncsu I see popping on the discussion list, I would assume that
the code for ncsu is not really a viable option for anyone using amber
these days.

2013/10/4 Jason Swails <>

> On Fri, Oct 4, 2013 at 3:00 AM, Fabrício Bracht <>
> wrote:
> > Hi all. Would it be possible to obtain the minimum free energy path of a
> > mechanism similar to the Grotthuss mechanism using umbrella sampling with
> > replica exchange. Let me try to contextualize this a bit more. For
> > instance, by stating that a proton exchange between three molecules (say
> > one hydronium and two water molecules) occurs in a concerted manner, that
> > does not mean that the distances between the exchanging protons relative
> to
> > their accepting oxygen atoms are the same. This depends a lot on the
> > surroundings, temperature...etc. So by simply using LCOD, for example,
> > between the exchanging pairs to describe these transfers, one would have
> to
> > try every combination (within reasonable known distances between pairs)
> in
> > order to obtain the path with the lower energy. Since I cannot use the
> > excess of charge coordinate to perform this type of simulation in amber,
> > nor can I use plumed (and its fancy variables customization properties)
> > with amber12, I thought that REMD might be a solution.
> > Does this sound like a reasonable proposal?
> >
> You can look at the adaptively biased MD options, which provide many
> similar collective variables as plumed. But as far as REMD is concerned --
> what type of REMD do you propose? T-REMD? Umbrella sampling-REMD? You
> can certainly use it, but I don't see how REMD solves any of the problems
> you have described. With T-REMD, I don't think you can see rapid proton
> diffusion at reasonable temperatures, whereas with umbrella sampling-REMD
> you still need to define your reaction coordinate the same way you would
> without replica exchange. Based on your statement about using LCOD, I take
> it that this approach is somewhat intractable.
> (Whatever you choose, though, requires using a flexible water model and
> turning off SHAKE, I would imagine, which further reduces the time step you
> can use).
> HTH,
> Jason
> --
> Jason M. Swails
> BioMaPS,
> Rutgers University
> Postdoctoral Researcher
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Sat Oct 05 2013 - 00:30:03 PDT
Custom Search