From: Bennion, Brian (
Date: Fri Jun 26 2015 - 20:32:01 CDT

From: Evandro Semighini []
Sent: Thursday, June 25, 2015 12:28 PM
To: Mayne, Christopher G
Cc: Bennion, Brian;
Subject: Re: vmd-l: Questions regarding fftk and Gaussian

Hello again,

I managed to answer some of my previous questions, but now I got into others, which I will appreciate if you could help me again.

The issues about Gaussian were in my computers/license. Using Gaussian 9 in another machine I managed to run the calculations in an appropriate time.

But, following the recommendations from Brian Bennion and Christopher Mayne (here and at NAMD list), and reading all I could find in VMD list about fftk, and also the Mackerell slideshow from the tutorial, I got stuck at the next step.

1- I need to run the Gaussian water interactions with the whole molecule or I can run only the ones from the region with high penalty from the PARAMChem ?
You could do this especially if the whole molecule is really floppy. You would spend forever relaxing the rest of the molecule after each move of the water molecule.

2- Considering that my molecule has 4 parts, A-B-C-D, I must split it in A, B, C and D parts or A-B, B-C and C-D to run the optimizations ?
The answer depends on how long you want to wait to finish your calculations and how easy it is to separate your molecule. Please look at the parameterization tutorial on the NAMD website, it may be dated but it walks you through breaking your molecule into pieces that look like previously parameterized molecules. You also need to be careful about the angles and torsions that span your cut points.

If the answer is in individual parts, to get the right charges I just need to sum the charges from the bonding ones and the hydrogens that replaces the "next atom" at the split ?
If the answer is in pairs, how I get the correct charges for the atoms common to the pairs ?
3- I must perform the optimization first in simulated annealing, save the log and then re-run the optimization with the downhill method, at FFTK ?
I made it first with downhill and the charges were the same from the Paramchem, but with sim_an, followed by downhill, the charges were different, some by more than 0.2.
Paramchem is an empirical engine using data already stored in the parameter files to guess the best parameters for your molecule. It has no idea of the energy landscape of your molecule which may be very different when compared to the QM properties of your structure. You need to look at the fit of the charges to your QM data. See a recent post from Chris on June 26th 2015.

4- Not all of my molecules got high penalties (some got none) from PARAMChem. To obtain a more standard result from all my MDs, I need to parameterize them with gaussian too ?
Not really. This is the point with paramchem, to minimize the space the you have to parameterize with each molecule. If you trust paramccem then use those parameters. Personally I would feel more confident if I just tried to parameterize one molecule with no penalties just to see that I am not making mistakes. Unfortunately I don’t have that kind of time nowadays.
Like I wrote earlier, if you have lots of molecules that are not related it may best to use amber and the gaff forcefield with antechamber.
If your compounds are structurally related for a majority of the molecule then you just have to parameterize the “new” parts for each molecule.
This is the charm of the charmm force field (transferability).

Sorry for some questions, but I got suspicious with easy answers.

Thank you all in advance.

Evandro Semighini

2015-05-19 17:07 GMT-03:00 Mayne, Christopher G <<>>:

Please note that parameterization is an advanced topic, even for experienced practitioners of molecular dynamics. I strongly suggest that before jumping into parameterizations of large, complex molecules, that you first work through parameterizations of smaller, simple molecules to gain a better understanding of the force field, and the more importantly, the established protocols for parameterizing for that force field. These protocols are not arbitrary, and deviations from them should be well thought out and justified to obtain reasonable and reliable results.

In addition to Brian's response I'll add that:

One of the key ways in which CHARMM is different from GROMOS and AMBER is the method for developing charges. The CHARMM protocol is founded on optimizing water interactions, which is critical to maintain compatibility with other CHARMM force fields (e.g., biopolymers, CGenFF). The charges returned by the QM software are computed by different methods that are (potentially) more meaningful to the QM calculation than any subsequent MM-based MD calculation.

Christopher Mayne

On May 19, 2015, at 12:45 PM, Evandro Semighini wrote:

Hello Brian,
Thanks again !
1- I read it and saw the difference between the force fields that was confusing me, but a friend of mine, that works with GROMACS and AMBER FF, told me to try the chelp charges.
2 an 3- God point, got it.
4- I got the files generated from fftk, it takes at least 3 hours for every run with one processor (with more, it crashes) in an Intel Core i3 2100 processor.
Unfortunatelly, the last advice will be the hardest part to get. =/
Thank you Brian !

2015-05-19 12:51 GMT-03:00 Bennion, Brian <<>>:
Please read Dr. Mayne's paper that describes ffTK, it will provide insight on question 1 (ie how charges are determined for the charmm forcefield).

Question 2. It is up to you to decide if the penalties given by<> are worth worrying about. If someone needed to repeat your work and they parameterized everything de novo, would you trust their results if they differed from your results?

Question 3. If you don't optimize the bonded parameters then you get what the cgenff forcefield supplies (or doesn't supply for that matter). The forcefield has NO idea that your compound needs to be "frozen" in that docked conformation, unless you tell it by giving it the appropriate parameters. Please realize that the docked conformation represents a theoretical complex at 0 Kelvin. Will you or others do the experimental studies at 0 Kelvin?

Question 4. It depends if you are running he calculation correctly or if your hardware/software is properly setup for these calculations.

These questions point out that you need to do some more reading and ask for guidance from your mentor/supervisor.

Brian Bennion
From:<> [<>] on behalf of Evandro Semighini [<>]
Sent: Tuesday, May 19, 2015 6:06 AM
Subject: vmd-l: Questions regarding fftk and Gaussian
Hello VMD community !
I'm new to NAMD and VMD and have some questions about the parameterization of the molecules I got from virtual screening assays.
I got a huge help from the NAMD mailing list and it leaded me to another kind of problems and doubts, regarding fftk and Gaussian, that I was unable to solve by myself/searching, so, here I am.
I was instructed by mrs. Crystopher Mayne and Brian Bennion to use ParamChem and then cut the molecules in pieces with molefracture to optimize them with fftk and Gaussian, and there's where I am stuck now.
As my molecules are sort of big, fftk creates more than 90 water interaction files to be loaded to Gaussian for each one, and I have more than 30 molecules.
1- Do I really need to run all of this water optimization in Gaussian or I can run just the geometry optimization, which calculates new charges ?
2- If no, can I run only the interactions for the highest penalized atoms or, there is other method with lower time cost ?
3- Do I really need to optimize the angles and dihedrals penalties, as my molecules are in the docked conformation ?
4- I never used quantum softwares before, so, is it normal this calculations take several hours to end ?

Thank you in advance for any help.
Evandro Semighini