Re: NAMD and usable FFs

From: Kenno Vanommeslaeghe (
Date: Thu Oct 16 2014 - 17:53:13 CDT

On 10/16/2014 05:41 PM, Francesco Pietra wrote:
> Again to CGenFF after a long while, I was terrorized by the complexity of
> symbols for atom types of organic molecules (protein ligands).

True, it's pretty complex indeed. That's exactly why we have encoded the
atom typing rules in the CGenFF program at . It gets it
right more often than I manage to do manually.

> Also, for
> ligands not too far away from purine bases, CGenFF performed quite
> naively, for example, giving a single Kb for bonds

I believe you might have misinterpreted the CGenFF output. By default, it
outputs only the parameters that are not already in the main parameter
file, so if it outputs a single parameter, that simply means all the other
parameters are in the main file. And I'm quite certain it doesn't use the
same bond parameter for C=O and C-C bonds; if it does, you should report
that as a bug. Conversely, if the problem is simply that the parameters go
against your intuition, then you probably should show they're misbehaving
in an actual simulation scenario.

> So, I built bonds, angles, dih,
> impr from antechamber. Then, apart from the problem of charmm-type partial
> charges,

As I've repeated countless times on this list, this is fundamentally
wrong. Bonded parameters (especially the dihedrals) only work in
combination with the 1-4 and longer-range electrostatic interactions
against which they were parameterized. If you're going to use antechamber
for your bonded parameters, then you HAVE to use GAFF atom types and
charges, and an Amber representation for the rest of your system (protein,...)

> I tried a couple of times ffTK
> just to be on an absolute basis, a very nice suite but extremely time
> consuming. Nothing good comes for little, however.

True. Relative to ffTK, how time-consuming would you say preparing,
setting up, running and analyzing an MD simulation is?

> Straightforward with normal proteins, the problems come
> with organic ligands, not to say with metalloproteins.

To such a vague accusation, we can only reply equally vaguely that our
software is more honest than most about the quality of the parameters it
produces, and that it will flat-out refuse to generate a model that would
likely be deeply flawed, reflecting our philosophy in releasing force fields.

If you're talking specifically about heme, you'll find the same heme model
you mentioned before in the CHARMM36 tarball.

> In that respect,
> which is nearly the norm for me, amber offers easier tools

No problem, we'll be just as happy if you choose to use amber. As long as
you don't mix them. ;)

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:56 CST