Please don't include the digest as if you're replying to it :-)
On the subject of your question, I would be happy to discuss how these
parameters came about, although I am not entirely certain of the protocol
myself. What I can tell you for the moment is that OPLS, in fact a
historical derivative of Amber which has long since grown into its own
thing, has over 5000 angle parameters for different chemical environments,
and those are all done at the B3LYP/6-31G* level according to their most
recent paper in JCTC. However, thanks to a paper last year by some of the
CHARMM devs we have some solid ways to derive angle parameters in their
respective contexts. The way to do this is actually to use two basis
functions: both parabolas, positioned with their respective minima on
either side of the minimum that you anticipate the angle to have. So, if
you think it should be in the neighborhood of 109.5, place 'em at 105 and
115. The sum of two parabolas is another parabola, and thus from the
coefficients you solve for each basis function you back out the stiffness
and equilibrium constant of the actual angle term. When you do this you
should hit the fitting program with sufficient data--the angle should be
seen getting more strained by stretching it too wide and squeezing it too
tight. The angles are not independent of one another nor of other
parameters in the system.
Fitting dihedral terms for X-ca-ca-X is a tall order, but in principle it
is the same as fitting ct-ca-ca-ct. What you must do is fill in the
wildcard X with actual atom types and collect data for many potential
energy surfaces involving -ca-ca- in the middle. Once this is done you can
fit a common set of parameters to best satisfy all anticipated atom types X
on the ends. It is a similar concept to applying the same dihedral terms
to atoms of the same types but with different partial charges in different
contexts. You must collect data for numerous relevant cases and choose the
consensus parameters.
If you'd like to go offline, I can tell you all about the development
capabilities in mdgx for fitting angles, torsions, and charges. I'm
looking to expand the particular charge derivation I have implemented in
there, but the charge FITTING code is general and can take whatever input
data you want to throw at it. The torsion and angle fitting module takes
everything as a single problem, and my collaborators and I are getting
ready to submit a manuscript on a new protein FF with some 750 dihedral
parameters and 60 reoptimized angle coefficients obtained from 265k QM
single point energies. We are including in the supporting info the entire
training set and mdgx input which runs in about half an hour to exactly
rederive the entire bonded parameter set.
Dave
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Mar 08 2016 - 22:30:03 PST
