[gmx-users] help on converting charmm/cgenff parameters to gromacs

[Mark Abraham]

On 6/04/2011 2:51 PM, Peter C. Lai wrote:
> Hello
>
> I am constructing a ligand for which I wish to use the new Charmm CGenFF
> parameters (a long aliphatic ketone).
>
> I am using Tom/Par's charmm36 lipid conversion as a baseline template for
> comparison:
>
> For reference, c36 lipid CTL3 atoms (in the case of POPC) map to CG331
> in CGenFF; they are both to represent alkane CH3 carbons. Same with
> CTL2 ->  CG321. In my case I am particularly interested in CGenFF's
> parameterization of a ketone carbonyl and oxygen: CG2O5 and OG2D3
> (without any associated carboxylic bias or bias from non-carbons found
> in other parameters. For example C=O in amino acid backbone appears in
> ffbonded.itp as 5.188e+05 for kB whereas Mark's script yielded a kB of
> 5.858e+05 for CG2O5=OG2D3, not to mention the bond angles should be
> different and especially the nonbonded interactions since we've replaced
> N with C and we have an extra H etc. etc.).
>
> My ligand for now is pretty simple:
>
>                        OG2D3
>                         ||
> CG331-CG321-...-CG321-CG2O5-CG331
>
> or, in c36 lipid/prot terms:
>                     O
>                     ||
> CTL3-CTL2-...-CTL2-C-C-CTL3
>
> How would I convert the relevant atoms in CGenFF's prm file to both
> dihedral and nonbonded interaction .itp for gromacs?

The standard CHARMM .prm files give an indication of how the parameters 
will be used, so it's just a matter of converting units and taking care 
of any constants.

>   I tried using Mark's
> perl scripts and it is giving me wrong LJ terms as well as not picking up
> any 1,4 interactions (not to mention it has no knowledge of Par's
> functype 9 for converting Charmm dihderals - we should no longer be using
> R-B functions).

This script predates dihedral functype 9, and implements a 
post-processing approach to deal with LJ, 1-4 and some dihedrals. I 
assume you haven't done the latter. In any case, the form of my 
solutions do not mesh at all with Par's solutions - forget about my scripts.

You seem to be only interested in about two atom types, so I'd expect 
you can convert all the relevant things by hand in under an hour.

> Finally, a slightly (un)related question:
> in ffnonbonded.itp, why does CTL2 get a charge of 0.05 when in reality we
> usually give it -0.18 for associated molecules in the rtp? Is there
> something I'm missing there? (CTL3 has -0.27 in both the itp and rtp files
> which makes sense to me).

Sounds like a glitch when the file was written - I guess some other 
carbon also has 0.05 and this was copied and pasted. In any case, the 
charges in ffnonbonded.itp are never used by pdb2gmx, because 
precedence-taking residue-specific charges are defined in the .rtp. If 
one were to use some other tool for generating topologies, I don't know 
whether charges missing from a [moleculetype] definition are looked up 
in ffnonbonded.itp.

Mark