VMD-L Mailing List

From: LOISON CLAIRE (claire.loison_at_univ-lyon1.fr)
Date: Tue May 24 2016 - 06:24:43 CDT

Dear Profs. Gumbart and Mayne,


Thank you very much for your rapid and precise answers !


I have read the nice article about the macrolides indeed,
it inspired me to cut the molecule into pieces.
I also take notice of the fact that I should reiterate the optimization. Thanks !


It would be great to have access to

some of the files of the dihedral optimization,

especially to get the MM final geometries,

and play with these !


Concerning the output from the initial parameters :

 maybe one could have a button "without optimization", or the possibility,

 during an optimization, to put the number of cycle to 0 (I did not try that).

And when you propose to save the plots,

maybe it is not complex to save the energy data into files at this point.

​

Best regards,

Claire





--------------------------------------------------------------------------------
Claire LOISON
Light and Matter Institute<http://ilm.univ-lyon1.fr/>
Theoretical Physical Chemistry Group
--------------------------------------------------------------------------------

  * 00 33 4 72 43 12 57
  * claire.loison_at_univ-lyon1.fr
  * ----------------------------------------------------------------------------
  *


________________________________
De : JC Gumbart <gumbart_at_physics.gatech.edu>
Envoyé : dimanche 22 mai 2016 01:57
À : Christopher G Mayne
Cc : LOISON CLAIRE; vmd-l_at_ks.uiuc.edu
Objet : Re: vmd-l: FFTK dihedrals and problematic intermolecular H-bonds

Thanks for the reference. If I understand the problem correctly, we ran into similar (or at least analogous) issues and developed an iterative procedure, described in the paper, to optimize charges using the molecule of interest along with a reduced version of it.

That being said, I don’t recall specific problems with the dihedrals afterwards as a result (not to say there weren’t plenty of issues anyway). The CHARMM force field itself is intended to be optimized in an iterative fashion, with second iterations and beyond targeted the MM-minimized geometry rather than the MP2 geometry. This may help with your specific issue.

Best,
JC

On May 20, 2016, at 11:11 AM, Mayne, Christopher G <cmayne2_at_illinois.edu<mailto:cmayne2_at_illinois.edu>> wrote:

Claire,

Thank you for the carefully thought out suggestions—we really appreciate them. I’ve replied inline below.

On May 20, 2016, at 4:29 AM, LOISON CLAIRE <claire.loison_at_univ-lyon1.fr<mailto:claire.loison_at_univ-lyon1.fr>> wrote:


Dear FFTK developpers,

The FFTK is a great help, thank you for your very nice job !

Here first small hints on what could be improved for the user in my opinion :

- in the torsion explorer and dihedral fitting :
adding small arrows to change the order of the QM log files.
When one loads several at the same time, they are in
order which are not always the best to investigate afterwards…

This should be pretty straightforward to do. I will also add this feature to the Torsion Optimization tab.


- in the dihedral results (energy curves).
Would it be possible to see the results of the
energy scan before any optimization (total MM energy from the input par) ?
Would it be possible to save energy curves into text files to compare
different approaches afterwards ?
Or alternatively : where are the files stored by FFTK during its working process ?

There are a couple of questions in here.
1) ffTK only writes out optimization information for the initial optimization. The location of that output s provided in the Input section (“Output LOG”). All of the refinement data is stored in temporary files that are cleaned up upon finishing the refinement routine; so, there’s currently no way to inspect that data.

2) I’ve long wanted to address the issue of plotting the MM PES for the starting point parameters, which would allow the user to hand tune the parameters during the final refinement, rather than relying completely on the optimizer (if desired). I think that the best solution is a second button just above “Run Refitting/Refinement” that would compute the MM PES based on the parameters provided in the box above. You could then compute this PES and subsequently run the actual refinement to compare the input and final refined output. I think that this would be nice addition that addresses your request.


- would it be possible to add a new function, after the model is done (after the dihedral fitting),
 where one could compare the QM and MM energies of abritrary
conformations (taken for example out of a G09 log file), without optimizing ?
My idea behind this question is to test the model on conformations which where not used before during the optimization, or to see the impact of parameters which are not considered before, of scans which are not the previous dihedral scans. I think for example about NBFIX parameters. The idea would be similar to the dihedral optimization, but without the optimization procedure. It would also be great to have the option to visualize and store several energy data for different par files.


I think I understand what you’re asking for here, but I don’t see a place in the current toolkit where this fits in. This would be a completely new feature that would take some time to develop. While I’ll keep it in mind, it will have to wait until I finish up some other development work.


And now questions linked to my difficulties with an optimization of a small organic molecule based on CGFF​. I had the impression that my difficulty emerge because of intermolecular H-bonds, between several close alcools, a bit like in glycerol or sugars. Therefore, I have cut the molecule into pieces to get most parameters when this intermolecular H-bonds are not present. When I use these parameters for the molecule I want, most of the dihedral scans are well reproduced, but their are a few isolated points where the MM energies are several kcal/mol lower.
 As far as I understood, these problematic cases appear
in the presence of intermolecular hydrogen bonds
with relatively long O...HO distance (2.3Ang) , which
seem too favorable in the MM model (by several kcal/mol !?).

I tried to play with the charges, but I was not able to improve.
I also tried some NBFIX, but this did not improve either.
Do you have any advise or strategy for that ?


I have not come across this problem in any of the parametrization that I’ve done; however, I know that fellow ffTK developer JC Gumbart and his postdoc struggled with this when parametrizing macrolides. You might look into their paper in J. Comput. Chem. (http://onlinelibrary.wiley.com/doi/10.1002/jcc.24043/abstractLXeK6tCS1H9ofZVJRtQ29r2yF2RvhfRE&s=KaiI-gX8tyGmRrUKJMB_V7LxpwECcE7jQlEW422e-Ts&e=>) for some ideas or contact them directly.


- During the MM energy calculation, are the MM configurations
forced to remain on the QM geometries or are they optimized
with the dihedral constrained ? If they are optimized, is it possible to investigate the
geometries of the MM dihedral scans to look for MM/QM discrepencies ?


We constrain any dihedrals that are being fit and perform a MM geometry optimization on any remaining degrees of freedom. We don’t currently output the optimized MM structures; however, I’ll consider adding a checkbox to the Advanced Settings to write this data out, as I see merit in what you’re after.

Thank you for any help ! Best regards,
Claire
​