Hi everyone:
I've been tearing my hair out over a problem of validation of new force fields
in Amber, and if there are good suggestions out there, please let me know.
A bit of history:
The early Amber ff files explicitly stated which improper torsions were to be
applied; (I think CHARMM still does this). When LEaP came along, the decision
was made to automatically assign an improper to central atoms with exactly
three bonded neighbors. The problem is that Amber treats impropers the same
as ordinary torsions, so that the order of the three surrounding atoms
matters. The decision was made that the order would be alphabetical in the
atom type of the non-central atoms, and it was carefully tested that this was
consistent with the earlier explicit definitions.
Atom types were pretty static in Amber for many years, but recently there has
been an expansion: the CX atom type for proteins (assigned to Calpha atoms),
the CI atom type in bsc0, and (now) four new atom types in the "chi-OL" mods
to RNA. People (including me) just rather arbitrarily assigned unused
2-character strings to the new atoms types, without thinking about its effect
on impropers.
The people doing the "chi-OL" mods chose C1 and C2 as new atom types, not
realizing that these conflict with the use of C1,C2,C3 for the ff03ua force
field. To avoid this problem, I changed C1->CF and C2->CE in ff10, then spent
a while finding the problem that caused. (Many thanks to Jason for his help
with all of this!) The *actual* difference is not important -- the improper
torsion terms keep atoms approximately planar with either order of surrounding
atoms, but "validation" (comparisons to energies from old simulations that use
a different ordering) is a problem.
I will try to see if there is some workaround for the present problem
(e.g. using C4,C5 in place of C1,C2). Longer term, we probably need to
migrate to an improper function that does not depend on atom order. And
maybe this note will prod people to check for problems like this when
adding new atom types.
As I said, comments or suggestions are certainly welcome.
...dac
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Mar 30 2011 - 07:00:03 PDT
