Re: [AMBER] A refresher on how non-bonding terms are represented in a topology file.

From: Jason Swails <jason.swails.gmail.com>
Date: Wed, 21 Sep 2016 20:48:31 -0400

On Wed, Sep 21, 2016 at 4:09 PM, Daniel Roe <daniel.r.roe.gmail.com> wrote:

> Hi,
>
> I'll do my best to answer these questions - corrections from others are
> welcome.
>
> On Wed, Sep 21, 2016 at 1:10 PM, Christian Schafmeister
> <drschafmeister.gmail.com> wrote:
> >
> > QUESTION #1: The excluded atom lists are indices into the atom tables
> in increasing numerical order?
>
> I think so, yes (related to #2).
>

​Yes.



> > QUESTION #2: The excluded atom list only includes indices (j) where j>i
> for atom_i?
>
> Yes.

>
> > QUESTION #3: Excluded atoms are 1-2 (stretch) and 1-3 (angle)
> interactions What about 1-4 (dihedral) interactions, are they in the
> EXCLUDED_ATOMS_LIST or not?
>
> Yes 1-4 interactions are excluded.
>
> >
> > QUESTION #4: Is there anything important I’m missing about non-bonded
> interactions? Magical negative indices or values or something that have
> special meaning?
>
> A negative index in NONBONDED_PARM_INDEX means HB term, otherwise it's
> a LJ term (mentioned in the docs).
>

​Currently, none of the codes permit both 10-12 and 12-6 terms exist for
the same system. I have seen prmtops where water atoms were assigned
elements in the 10-12 arrays that were set to zero, which I think was a way
of "marking" them for SETTLE. But that convention is no longer used. (I
came to that conclusion based on a number of comments across different code
bases).

A few other gotchas:

1. The 1-4 nonbonded list is constructed from the dihedral lists (dihedrals
with H and without H). To indicate that a pair of atoms should be included
in the 1-4 nonbonded calculation, exactly one of the terms needs to have
its third atom index be positive. All others (such as multi-term torsions
and one of the 2 distinct torsions in a 6-member ring that have the same
end atoms)​ need to have its third atom index be negative. Likewise, any
pair of atoms that are ends of a torsion *and* either an angle or bond
(true for 5- and 4-member rings), all of those torsions need to have their
third index be negative (since the 1-2, and 1-3 exclusions trump the 1-4
exception rules). All impropers need to have its fourth atom index be
negative (and, since there is a central atom and no pair are ever separated
by more than 1 bond, the third is always negative too).

A consequence of this is that if a torsion involves the first atom (whose
index would be "0"), the ordering must be reversed if that atom appears
third or fourth in the torsion list if it's either improper or an omitted
term.

2. The last quirk I recall about the exclusion list is that for atoms that
either have no exclusions (like ions) or whose excluded pairs have a
smaller index, they are assigned a *single* exclusion to the non-existent
atom 0 (since the prmtop uses 1-indexing everywhere). Extra points also
have weird rules. They have the same exceptions as their parent atom.

Amber programs rebuild the exclusion list for PME calculations, but use it
for GB calculations. If you want to test assignment of the exclusion list,
you will either need to stick with non-periodic test systems (which don't
support virtual sites) or use ParmEd's OpenMM integration (which *does* use
the exclusion list in the prmtop).

The PDF file you found is a brain dump I committed to (what has come to
replace) paper several years ago. Another source of weird quirks are the
ParmEd comments, but the ones I listed are the big ones I remember learning
about.

The OFF file is a whole 'nother can of worms :). I also documented that
format awhile ago: http://ambermd.org/doc/OFF_file_format.txt, which is
just a somewhat refined and condensed summary of tleap code, comments,
commit logs, and experimentation :).

​The parsers and writers in ParmEd are fairly mature, and have undergone
several iterations of finding and fixing bugs for weird corner cases. It
may be a helpful resource if you can grok the code there (
https://github.com/ParmEd/ParmEd/tree/master/parmed/amber). Support for
the CHARMM force fields would also be cool :).

All the best,
Jason

-- 
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Sep 21 2016 - 18:00:03 PDT
Custom Search