Re: Protein-ligand simulation

From: Kenno Vanommeslaeghe (
Date: Mon Jun 24 2013 - 21:51:34 CDT

Hi Guys,

Allow me to add some background info:

There is parameter *assignment*, which takes less than a second and during
which the computer basically makes an educated guess at what would be good
parameters, typically based on similarities between the ligand and
existing molecules in the force field. For the specific case of CHARMM, I
know of 3 interfaces that do so:
        * SwissParam
        * the CGenFF program at
        * MATCH from the Charlie Brooks lab
Out of these three, I recommend *against* SwissParam because it's based on
an unfortunate mixing between MMFF94 parameters and CHARMM parameters, as
discussed on a few occasions on the CHARMM forums. Also, since I wrote the
CGenFF program, I'm obliged to say that CGenFF is da best(*) :)

Then there is parameter *optimization*, which typically takes the output
of the above parameter *assignment* as an initial guess, and refines it
using QM and experimental target data. FFTK is an example of the latter -
there exist others as well.

Putting this together in a NAMD context, present-time best practice would
probably be to feed your ligand to the CGenFF program at ,
then if there are high penalties (which is more often the case than not)
feed the results to FFTK. Among other advantages, this course of action
allows you to avoid the tedious and error-prone "System Preparation II:
Setting NONBONDED Parameters By Analogy" step in the FFTK tutorial.
Indeed, the CGenFF atom typing rules have become unusually complex, with
many exceptions, and I currently have more faith in the CGenFF program's
ability to assign correct atom types than in any human being (including
myself and Alex MacKerell).



(*) Just kidding. To be serious, there does not yet exist an independent
comparative study on full-sized drug-like molecules showing the strengths
and weaknesses of CGenFF vs. MATCH, but I'm looking forward to it. What
speaks in favor of CGenFF right now is:
        * it's the reference implementation of the CGenFF atom typing rules
        * broader support for linkages between functional groups

On Jun 24, 2013, at 7:55 PM, JC Gumbart wrote:
> Clearly FFTK. :)
> On Jun 24, 2013, at 2:15 PM, James Starlight wrote:
>> Chris, thanks you! I'll be very thankful for your tutorial!
>> By the way what is the current most useful way to obtain charm parameters for small ligand-like compounds (starting from its full atomic pdbs with hydrogens) ?
>> With Gromacs (using charm36 force field) typically I use swiss-param server which also produce PRM files in addition to the gromacs topology.

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:23:22 CST