From: Brian Radak (
Date: Tue Nov 13 2018 - 08:37:37 CST

Despite the wide interest, I would say this is still a largely unresolved

Indeed, GAFF is, in my recollection, a rather deterministic procedure.
CGenFF at least gives penalty scores when it believes it is doing poorly,
but there really isn't much recourse in that case other than to do a
detailed quantum chemical study of the molecule (e.g. using FFTK). To the
best of my knowledge, using a covalent ligand immediately jumps your
workflow to the detailed quantum chemical study, since neither GAFF nor
CGenFF are really prepared for that.

There was a recent paper on the parameterization of cobalamines from the
Gumbart Lab - maybe that will offer some best practices? I know that they
used FFTK. At the very least you could probably get a hold of their


On Tue, Nov 13, 2018 at 2:14 AM Francesco Pietra <>

> Hi Brian:
> I understand your point, which was also my suspicion, though not clearly
> declared. Covalent ligands are no rarity, actually pharma is paying much
> attention to them, under the hope that they reside longer than non-covalent
> ligands at the active site.I remember vaguely of some simulation work with
> cytochrome c. If I am correct, that was a case of covalent ligands. But who
> did that, and with which ff?
> It seems to be easier with amber ffs, since gaff is compatible with them,
> although gaff is speedy to provide a ff for the ligand as if it were a
> straightforward affair.
> cheers
> francesco
> On Mon, Nov 12, 2018 at 7:19 PM Brian Radak <> wrote:
>> Francesco,
>> I'm not sure that what you are attempting is well-advised. I don't think
>> CGenFF was ever intended for covalent linkage to the protein force field
>> and I don't think there is any automated path to do so. There is *some*
>> correspondence between CHARMM36 and CGenFF types and you could notionally
>> convert CGenFF interactions by matching, but my understanding is that this
>> has never been suggested by the MacKerell group or others.
>> You will probably have to find a detailed procedure from the literature
>> to follow. Unfortunately I have no experience with covalently linked
>> ligands, so I have no tips to offer there.
>> HTH,
>> BKR
>> On Mon, Nov 12, 2018 at 12:51 PM Vermaas, Joshua <>
>> wrote:
>>> If residue 1 in the patch is supposed to be HSP, why are you setting the
>>> type/charge of a likely non-existent atom?
>>> "ATOM 2NE2 NG2R52 -0.133"
>>> This should probably be ATOM 1NE2, shouldn't it?
>>> -Josh
>>> On 2018-11-12 02:02:08-07:00 wrote:
>>> Hi:
>>> I am trying to get bonding between protein HSP NE2 and an organic ligand
>>> carbon, while imposing CGenFF atom types on both sides.
>>>> PRES ZAHI 1.449 ! patch for ligand to HSP bonding
>>>> ! Patch must be 1-HSP and 2-ZPN
>>>> ! use in patch statement
>>>> ! follow with AUTOgenerate ANGles DIHEdrals
>>>> command
>>>> GROUP !
>>>> ATOM 2C11 CG311 0.332 !
>>>> ATOM 2C10 CG314 0.199 !
>>>> ATOM 2H15 HGA1 0.09 !
>>>> ATOM 2C9 CG2D2 0.127 !
>>>> ATOM 2C8 CG2O5 0.333 ! CD2--NE2--CE1
>>>> GROUP ! /
>>>> ATOM 2CD2 CG2R51 0.219 ! C11--C10--C9--C8
>>>> ATOM 2NE2 NG2R52 -0.133! |
>>>> ATOM 2CE1 CG2DC1 0.349 ! H15
>>>> BOND 2C10 1NE2
>>>> IMPR NE2 CD2 CE1 C10
>>>> IMPR NE2 CE1 CD2 C10
>>> However, the new atom names are only accepted for the ligand, not for
>>> the protein, which conserved the original atom types CPH1 NR3 CPH2 for CD2
>>> NE2 CE1, with bonding between C10 (CG314) and NE2 (NR3). pdb file shows
>>> zero coordinated for CD2 NE2 CE1 on the ligand residue.
>>> All that because I am trying to avoid parameterization for unusual
>>> bonds, such as between CG314 and NR3, although I understand that with atom
>>> type NG2R52 for NE2 a cascade of problems would arise on the protein side.
>>> So, which is the best way to associate CGenFF to charmm36?
>>> thanks
>>> francesco pietra