VMD-L Mailing List
From: Bryan Roessler (roessler_at_uab.edu)
Date: Tue Oct 20 2015 - 15:34:48 CDT
Yes, I do believe that is a proper methodology. Just remember that as you
delete bad water interactions, there is less QM data to guide the
optimizer, so if you have a lot of waters flying away it might be
worthwhile to reoptimize the geometry and rerun the water interaction
calculations to see if you can resolve some of the steric problems.
*Bryan Roessler | Graduate Research Assistant*
UAB | The University of Alabama at Birmingham
Knowledge that will change your world
On Mon, Oct 19, 2015 at 10:59 AM, Chitrak Gupta <chgupta_at_mix.wvu.edu> wrote:
> Hi Bryan and Christopher,
> Thanks for this discussion, cleared up some of the doubts I had.
> So, to make sure I got this right......I should start by optimizing water
> for any atom that should in principle have water interaction, correct? And
> then if they are flying out, just not include the log file of those waters
> in further calculation, is that what you suggested?
> Best regards,
> On Mon, Oct 19, 2015 at 10:50 AM, Mayne, Christopher G <
> cmayne2_at_illinois.edu> wrote:
>> Apologies for a late reply; I just returned from international travels
>> this morning.
>> Your own suggested answer is quite on target, and it sounds like you have
>> a pretty good handle on what is going on at this stage of the
>> Christopher Mayne
>> On Oct 19, 2015, at 4:45 PM, Bryan Roessler <roessler_at_uab.edu> wrote:
>> I'm going to take a stab and try to answer my own question based on some
>> further observations.
>> I believe that it is necessary to remove any waters that do not reach a
>> proper minimized state (for instance if they fly out of the Gaussian
>> simulation). It is best to try and include as many water interaction target
>> files as possible for the fragment (even for atoms that have predefined
>> charges), but if some of them skew the optimization results (for instance,
>> if they are not near the energy objective in COLP), then it is best to
>> reduce their weighting in the calculation or to remove them altogether.
>> Following these guidelines I have been able to get realistic optimized
>> charges near the CGenFF predictions.
>> If I am in error, please let me know.
>> *Bryan Roessler | Graduate Research Assistant*
>> UAB | The University of Alabama at Birmingham
>> Knowledge that will change your world
>> On Fri, Oct 16, 2015 at 10:01 AM, Bryan Roessler <roessler_at_uab.edu>
>>> Thank you very much for the reply. I should have included a second,
>>> related question in my original query: since I am not going to include the
>>> predefined atoms in the charge optimization step, is it still necessary to
>>> calculate the individual water interactions for those atoms and include
>>> them as input in the optimization?
>>> Because it is necessary to include the known atoms from the bonded
>>> residue, the complexity of the fragment also increases (~35 atoms total).
>>> As a result, I've found that some of the waters will fly out of the
>>> Gaussian optimization, presumably due to steric quantum interactions with
>>> the surrounding atoms. I know that the preferred solution is to reduce the
>>> complexity of the fragment by breaking it down further into its molecular
>>> components (between aliphatic carbons) but I've found that the partial
>>> charges that are assigned using this method do not reflect the
>>> intermolecular charges well (possibly because the fragment contains an
>>> ester). I think it is better to keep the fragment together as long as it is
>>> OK to remove some of the errant water interaction calculations from the
>>> optimization. What I don't entirely understand is if the water interactions
>>> can be removed from the initial calculation (and thus not reflected in the
>>> single-point energies of the entire fragment? I don't know) or need to be
>>> removed only from the optimization step so that those atoms are still
>>> included in the single-point calculations of the entire fragment.
>>> In my trial and error, I have tried to calculate only the water
>>> interactions for the atoms without existing FF charges, however when I
>>> perform the optimization using this method, the charges on the atoms tend
>>> to drift towards an integer charge.
>>> Thanks again,
>>> *Bryan Roessler | Graduate Research Assistant*
>>> UAB | The University of Alabama at Birmingham
>>> Knowledge that will change your world
>>> On Fri, Oct 16, 2015 at 2:50 AM, Mayne, Christopher G <
>>> cmayne2_at_illinois.edu> wrote:
>>>> Apologies for a delayed response; I have been traveling internationally.
>>>> Yes, it is generally a good idea to include enough of the known atoms
>>>> to cover the connection between known and unknown parameters. ffTK is
>>>> designed so that it is easy to optimize a subset of parameters — in this
>>>> case, retain the known parameters and optimize only the missing ones.
>>>> For the specific case of charges, the typical method used is:
>>>> 1) Set any “known” charges in the PSF file.
>>>> 2) When setting the Charge Constraints (Opt. Charges -> Charge
>>>> Constraints -> Charge Group) simply remove the known atoms from the
>>>> constraints box.
>>>> —> This process will optimize any atoms contained within the charge
>>>> constraints while fixing the charge on the atoms that are not included in
>>>> the box.
>>>> When analyzing the optimization output, the raw objective function is
>>>> not particularly informative. The COLP tool (available from Opt. Charges
>>>> -> Results -> Open COLP) allows users to deconvolute the objective function
>>>> and inspect each of the contributing terms. This allows you to judge which
>>>> terms contribute the most to the objective function, and to identify
>>>> problematic atoms.
>>>> Christopher Mayne
>>>> On Oct 13, 2015, at 9:08 PM, Bryan Roessler <roessler_at_uab.edu> wrote:
>>>> When parameterizing novel molecular fragments that will eventually be
>>>> bonded to a residue with an existing forcefield, it is necessary to
>>>> incorporate some of the atoms near the bond from the residue with existing
>>>> forcefields that contribute to the dihedral, angle and bond parameters in
>>>> the QM/MM optimization scheme.
>>>> Is it then suggested that when performing the charge optimization step
>>>> in fftk, I maintain the charges on those overlapping atoms as they exist in
>>>> the CHARMM ffs as best as possible? In other words, should I create an
>>>> upper and lower bound (so that the optimizer still has some 'play') that
>>>> are very near the charge values on those atoms provided in the existing
>>>> forcefield? Or should I let all of the charges optimize independently of
>>>> the existing forcefield charge parameters?
>>>> When I allow all of the charges to optimize without strict bounding, I
>>>> can attain a total objective that is very near zero after several iterative
>>>> optimization passes. However, when I loosely 'fix' the charges of some of
>>>> the overlapping atoms that already exist in the forcefield, the charge
>>>> optimization objective values are not nearly as good (which is to be
>>>> expected as the degrees of freedom are reduced). However, in some cases the
>>>> charges on the overlapping atoms can be quite different than the existing
>>>> charge values in the the CHARMM ff if I let them optimize without
>>>> Thanks for your help in advance,
>>>> *Bryan Roessler | Graduate Research Assistant*
>>>> UAB | The University of Alabama at Birmingham
>>>> Knowledge that will change your world