[gmx-users] Manual refinement of ATB topologies ?

[Justin Lemkul]

On 12/19/16 7:50 AM, Sim gmx wrote:
> Thank you for your answer.
>
> "Unfortunately", this molecule has a peculiar structure with a 5-atoms
> cycle (including a nitrogen atom) directly bound to a C=O itself bound to a
> CH involved in a double bound. I guess that this nearness between the
> groups should lead to some "hardly predictable" charge distribution within
> the molecule. Hence, if I submit for instance only the 5-atoms cyclic part
> to ATB and take the C=O parameters from an existing topology, I guess I
> will have a hard time to merge the two parts, am I wrong ?
>
> If I don't get you wrong, my 'instinctive behavior' shares some
> similarities with what you suggest (replacing, wherever it is possible, ATB
> parameters by 'known parameters'). But if I don't  submit the whole
> molecule to ATB, then I don't know how to get "reliable" atomic charges ?
>

This is common to all additive force fields.  You need a suitable model 
compound, one that includes linker portions that can be merged with neighboring 
functional groups by combining charges and applying/modifying known dihedrals. 
What I would do is try to parametrize:

(ring system)-C=O-CH=CH-CH3

and whatever might be a suitable flanking group for the ring (e.g. methyl or 
ethyl, etc) if it is in the middle of the acyl chain.  There may be partial 
charges on those neighboring methyl/methylene groups.  That would be normal. 
But putting partial charges on UA carbon atoms multiple bonds away is not 
intuitive, given the philosophy of the force field.  I would assume 
positive-positive repulsion would perturb the bilayer, unless the LJ mask the issue.

> You underline another important thing to consider: the choice of the right
> forcefield. Until now I've been working with berger lipids as forcefield
> for my bilayers (initially following one of your tutorials, by the way
> thanks a lot for this very helpful work !) in combination with small
> gromos53a6 molecules. Here, since my molecule includes a large acyl chain,
> it could be non ideal to use gromos53a6 parameters while the lipids with
> which it should interact are parametrized with berger lipids. Maybe berger
> - berger non-bonded interactions would be better for these mainly
> hydrophobic interactions ? Nevertheless I don't see how I could create a
> berger topology for such a peculiar molecule, especially because I don't
> know any ATB-like for this FF.

Berger lipids were derived from old GROMOS parameters and some aspects of OPLS, 
so don't get too hung up on whether or not interactions are Berger-Berger or 
Berger-GROMOS.  They're compatible.

> Maybe another force field could be used for both the bilayer and my
> molecule (with the help of a reliable ATB-like website) ? Or would it be
> better to use gromos53a6 lipids instead of berger lipids ?
>

I do everything with the CHARMM force field nowadays.  Parametrization is 
straightforward and the CGenFF server parametrizes small molecules and model 
compounds easily, and they can then be converted to GROMACS format with a script 
from the MacKerell group website.  The parametrization theory and protocol for 
CHARMM is published in extensive detail, as much or more so than any other force 
field out there.  The lipid force field reproduces many experimental properties 
well.  For those reasons, I think it is an optimal choice in a situation like 
this.  Of course, that comes at the price of more expensive simulations 
(all-atom with required force-switching) but that's a price I find worth it.

-Justin

-- 
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================