[gmx-users] molecule parameterization in charmm36 using GAAMP server

Justin Lemkul jalemkul at vt.edu
Fri Nov 18 20:27:54 CET 2016

On 11/17/16 12:36 PM, Mohsen Ramezanpour wrote:
> Thanks David and Justin for your comments.
> David: since I am interested in lipid lipid interactions, and lipids are
> well-parametrized in charmm36, I decided to not parametrize whole lipid in
> CGenFF.
> As I understood and Justin pointer out, CGenFF is good for small molecules
> (drug-like) which there is a few of them in the system not a lipid system.
> Justin: make sense. I thought so, but, here is what I did and what I was
> interested in:
> I have a new lipid which the head group is a bit strange (it has a
> dioxolane ring in it) and I could not find any similar part in charmm36 by
> analogy.
> I could get the parameters for the rest of molecule from charmm36 by
> analogy, though. Even for smaller pieces of head group.
> To get an initial idea about this ring, I used CGenFF and got some
> parameters for the ring part and dihedrals involved this ring. I used the
> charges from Swissparam as the first approximation for the whole lipid.
> Fortunately, CGenFF gave me ~ good parameters for some dihedrals, but not
> good enough for two of them (based on Penalty values of higher than 50).
> Thus, I picked some parameters (some atom types and dihedrals) from CGenFF
> output files. I could pick because CGenFF will find these parameters based
> on analogy, in a similar way to how I found the parameters for the rest of
> lipid. Besides, there are parameters for bonded and non-bonded interactions
> between CGenFF and Charmm36 atom types.
> So, I think till here is fine (consistent) and makes sense. Do you agree?

The parameters for ethers in CHARMM should cover this reasonably well.  Dioxane 
is also a standard model compound in CGenFF, so you should compare the sanity of 
whatever parameters you have against ethers and pre-built dioxane.  If there are 
large parameter penalties, that means there are other groups causing problems 
and you'll have to provide more information for me to be able to give better 

> Doing simulations, I found that charges and two dihedrals (the ones with
> high penalties from CGenFF) cause problems. So, I decided to do QM
> calculations on 1) charges, and 2) only two dihedrals not all of them.

Targeted refinement is reasonable, yes.

> GAAMP seems to be a good option for this. Although I wanted only two
> dihedrals, but for calculating charges, I think it is logical to submit the
> whole head-group to GAAMP. Doing this, GAAMP gave me parameters for the
> desired parameters with some modifications on parameters I got from analogy
> (either by myself or suggested by CGenFF).
> This is why I rather to keep parameters I found by analogy and just pick
> the parameters for charges and two dihedrals.

This seems pretty reasonable.  The challenge in any additive force field is 
merging independent building blocks into a robust model for the full compound. 
Specifically targeting problematic dihedrals is what we do all the time, just be 
sure that any time you modify anything (charges, atom types, etc) that you're 
iterating back to the QM target data to be sure that the conformational 
energetics are correct.  For dihedrals, this means 1-D potential energy scans 
for each of the dihedrals.


> Regarding all of these, do you think it is better to take all the
> parameters from GAAMP server? (GAAMP seemingly has also suggested new
> parameters for TIP3P which I think better to stick to the original water
> model, the one charmm36 was built on ).
> Any suggestion is highly appreciated in advance.
> Cheers
> Mohsen.
> On Thu, Nov 17, 2016 at 6:05 AM, Justin Lemkul <jalemkul at vt.edu> wrote:
>> On 11/17/16 12:56 AM, David van der Spoel wrote:
>>> On 17/11/16 01:24, Mohsen Ramezanpour wrote:
>>>> Dear Gromacs users,
>>>> I used GAAMP server to get the parameters for a small molecule (this is
>>>> part of a larger molecule).
>>>> This server optimizes the charges and all the soft dihedrals in an
>>>> iteration way (consistent with charmm36 and CGenFF philosophy). The
>>>> output
>>>> files also include modified angle, bond, and all soft dihedral
>>>> parameters.
>>>> This means, all the parameters are dependent to each other and I have to
>>>> use all of them together in my molecule. However, I am not sure about
>>>> this
>>>> claim.
>>>> My question is then, weather or not it would be OK to ONLY pick the
>>>> parameters for "charges" and "ONE of those dihedrals" NOT the suggested
>>>> optimized parameters for atom types, angles, bonds, and other soft
>>>> dihedrals.
>>>> This is because, when I made the initial .itp file for my molecule, I
>>>> could
>>>> find bond, angles and soft dihedral parameters by analogy to existent
>>>> molecules in charmm36 ff. So, I want to trust on those analogy-based
>>>> parameters for now. The only parameters I was in doubt about were
>>>> "charges"
>>>> and "one of those dihedrals".
>>>> Please let me know your opinion. Thanks in advance
>>> Why not use the official server
>>> https://cgenff.paramchem.org/
>> For a standalone molecule, this is indeed the easier option.
>> There is a distinction between GAAMP and the CGenFF server that may be
>> worth mentioning.  GAAMP completely parametrizes a molecule, therefore
>> assuming that no parameters for any of that molecule actually exist, so it
>> starts with doing QM, deriving charges, fitting torsions, etc.  It does all
>> the work that we do manually (or semi-automatically) when building a force
>> field.  This is a good approach if you have something really wild that
>> CGenFF won't be able to handle, because CGenFF assigns parameters based on
>> analogy, applying a best fit for anything that it doesn't explicitly
>> cover.  So each has its place, but even the output topology from GAAMP
>> isn't necessarily a "final" topology as additional refinement may be
>> necessary.
>> To the OP's question, no, you can't just pick and choose pieces of the
>> topology that you want to use and call it done.  Force fields are
>> self-consistent entities, and things like charges, atom types, and
>> dihedrals (which really are just correction factors for 1-4 interactions
>> due to the failure of nonbonded forms to universally treat inter- and
>> intra-molecular interactions) are all interrelated.  You can't necessarily
>> just take what you like and leave the rest, especially with dihedrals,
>> which are poorly transferable when making changes to the 1-4 atoms.  The
>> dihedrals depend on the bonds and angles, because those affect the internal
>> geometry, which affects 1-4 distances, which affects the strength of the
>> interaction between those atoms, which in turn determines the strength/form
>> of the dihedral necessary for that torsion.
>> -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
>> ==================================================
>> --
>> Gromacs Users mailing list
>> * Please search the archive at http://www.gromacs.org/Support
>> /Mailing_Lists/GMX-Users_List before posting!
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>> send a mail to gmx-users-request at gromacs.org.


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


More information about the gromacs.org_gmx-users mailing list