Re: [AMBER] Can we combine amber and gaff atom types?

From: David Cerutti <>
Date: Wed, 2 Nov 2016 16:44:42 -0400

^ Agree with Carlos, the interface between parameter sets should be
minimal. This is what I was getting at further down in my novella--take
ff14sb atom types for everything up to perhaps the sulfur on the side
chain. Now if we can retain the charges up to CB (type 2C) we can ensure
that the ff14sb backbone behavior (which is where the most effort has been
done) stays put. That's what we can focus on now. First, reference
amino12.lib in the $AMBERHOME/dat/leap/lib directory and mirror the charges
for ACE, NME, and the modified CYS up through CB. As needed you may
distribute the excess charge this imparts to the sulfur and the first
carbon on the pyrrole to get the net charge back to (-1).

To get the best parameters for ff14sb you will need to refit charges on the
pyrrole group and sulfur while retaining the other charges that correspond
to atoms ff14sb does describe. I wouldn't recommend trying to map the
pyrrole atom types to ff14sb equivalents: that thing has chemical groups
that aren't in proteins, which puts you in a different ball park. As soon
as you need new atom types, they might as well be called "plain", salted",
and " honey roasted," so the goal here is to invest an appropriate amount
of effort to make something compatible with the rest of your protein. The
central dogma of ff14sb is "Hartree Fock 6-31g* charges, with bonded
parameters to mimic a vacuum phase MP2 / cc-pvTZ potential energy surface"
and that's what we'll do.

One thing I didn't realize when I first replied is that your modified amino
acid is BIG. The first capability we need is to be able to make plausible
conformations of your nonstandard residue. Cysteine covalently bonded to
not just one ring system but a macrocycle of four! This is going to create
a challenge, but what it means is that you may want to trim away the excess
before trying new parameter development. *My advice would be this: take
bond and angle parameters from our libraries, decide on charges for the big
system, then create a stripped-down, tiny system and try to develop the
torsion parameters for the pyrrole :: ethanethiol interface.* In what
follows I'm providing detailed instructions to carry out that idea.

First, take bond and bond angle parameters straight from ff14SB or GAFF,
using Carlos's logic: these parameters are derived prior to charge fitting
and can, in most protocols, be taken as given. They're also more
consistent across different atom types ({2C, CT, 3C}-CT-N, etc.)--the
different atom types largely come from our efforts to get increasingly
accurate torsions in slightly different chemical situations, but when we do
that we typically just clone the bond and bond angle parameters from their
root types (2C, 3C, and CX are all from CT). Just find the best matches
you can based on your chemical intuition, probably based on the ff14SB
methionine residue, and apply the stiffness and equilibrium parameters to
the new atom types at the pyrrole :: SG-CB interface.
Next, charges: if after reassigning the charges of the Cysteine up to SG,
you only need to change the charge of that sulfur by 0.1e to maintain the
integer net charge, I think you're set. (The charge on sulfur is never
well resolved by our methods--the thing's got quadrupoles that our
nuclear-centered charges simply can't do very well, so R-S-R' sulfur atoms
in methionine and other situations always get assigned very small charges
because the least squares solution is to give it no significant monopole,
forfeiting whatever higher order multipoles would be better.) Otherwise, if
you can find out how to make the RED server give you the charges you need
for the SG and nearest atoms on the tetrapyrrole, you're set. If neither
of those solutions works, I can hit it with mdgx and see about getting you
a reasonable charge model directly from HF/6-31g*.

Finally, the torsion parameters: for this, you should excise one pyrrole
ring from your macrocycle, put CH3 blocking groups on either end where the
macrocycle got cut, and attach -S-CH2-CH3 to the other side. Then, do the
same trick of assigning ff14SB atom types to the S-CH2-CH3 atoms (i.e. use
2C for the carbon of CH2, CX for the carbon of CH3, and H1 for all
hydrogens on CH2 or CH3) and the atom types of your larger Cys-Pyrrole
system for the pyrrole ring. Once you have that small molecule, just
assign it the standard AM1-BCC charges (again, I'm counting on the sulfur
getting a pretty small, flat charge asssignment). And keep the bond and
bond angle assignments from your larger Cys-pyrrole, but set the torsion
amplitudes involving the sulfur to zero (you're going to make a new frcmod
for this representation of the linker system in the same way it seems you
have for your Cys-pyrrole already). Now you've got a model system which
can move in the ways you need to see the joint between your larger Cysteine
and pyrrole moving. Once you've got that, I think the easiest solution is
to use the new mdgx routines to generate a handful of conformations (about
60 I'd guess), and computations at the MP2/cc-pvTZ level will take about an
hour each on a single CPU to get their single point energies--an afternoon
with a couple of machines. With the conformations of your test system and
their single point energies in hand, it is a five minute exercise to fit
the torsion parameters appropriate for the links between the ff14SB and
GAFF components of your system.

Last, you collect all of these bond, bond angle, and torsion parameters
into a frcmod and take the charges and ff14SB/GAFF atom types in your
.prepi or .lib file for the full cys-pyrrole, and you're ready to do your

Keep in touch, I'll do what I can to help.

AMBER mailing list
Received on Wed Nov 02 2016 - 14:00:03 PDT
Custom Search