Re: [AMBER] About restraint

From: Thomas Cheatham <>
Date: Tue, 17 Apr 2012 21:30:03 -0600 (Mountain Daylight Time)

AMBER is not set up for this type of calculation (i.e. where we want to
fix or restrain part of the system in order to reap huge savings in
computational cost). Others may have greater insight, but it is not clear
to me even if *other* codes are able to significantly increase performance
by "fixing" or treating as static part of the system anymore.... In the
past, tricks could be applied, however with Ewald/PME (or current implicit
solvents) effectively you have to go over all the atoms regardless and
then correct for the fixed parts. There is no big savings. (If there
were, we would have likely implemented it!).

In AMBER, if you "fix" atoms (for example with IBELLY) still all the
forces are calculated, then zeroed. For restraints, still you traverse
over all atoms plus then you have the added expense of the restraints...
The only way around this is like with pseudopotentials in QM, treat part
of your big molecule as a large sphere with no molecular detail. You
could perhaps delete all the residues within a sphere in the middle
(assuming of course there was a way to maintain the connectivity outside
this) and put in a large sphere. Perhaps with dummy atoms to maintain
connectivity (where the chains were broken). I think you could do a
couple thesis projects on this. Then the hard part would be convincing
the reviewers and community that this is an accurate representation of
your large molecule.

> I'm not sure to what extent that is possible without making the simulation
> meaningless. The intramolecular forces for your "big" molecule are almost
> certainly critical for it's structure, and it doesn't have just 1 single
> structure, but an ensemble of structures that will contribute differently
> to the interaction with your "small" molecule, and thus must be
> considered. That being said, if you really want to continue, I think you
> can fix the positions of certain atoms, using a restraint mask (or
> something like that) and thereby not have to calculate those interactions.

I agree on the ensemble of conformations.

> If you are doing PME with periodic boundary conditions, the electrostatic
> cutoff will already help you by not calculating the direct electrostatic
> interactions between distant parts of your large molecule.

Yes, in some sense, however there still is a significant cost in the
direct space interactions and the costs grow for both the direct and
reciprocal with number of atoms. However, the repliers point maybe, try
it out with all the atoms and see if the cost is prohibitive. If it is
prohibitive, you are out of luck since there are no easy tricks to get
around this...

My advice would be (a) either run the whole thing, or (b) design a smaller
structurally reasonable interface of the large macromolecule and use this
to probe ligand-molecule interactions.


AMBER mailing list
Received on Tue Apr 17 2012 - 21:00:03 PDT
Custom Search