[gmx-users] HF/6-31G** ESP derived charges to replace PRODRG assigned ones

[Ran Friedman, Biochemisches Inst.]

Dear Josmar,

You haven't written which force field you plan to use. For OPLS and AMBER 
QM-based optimisation should be fine. In Gromos, the FF was developed with 
the aim of reproducing experimental results and I'm not sure if you can find 
a better solution than examining other residues with the same chemical 
moieties or use the same approach as reported in the relevant manuscripts. 
Some software packages can also be used - these are mostly proprietary and 
not so easy to use.

Once you derive the parameters, it's a good idea to make some test runs of 
the ligands and see if they behave as expected before you actually run a 
simulation with the protein. For example, if a conjugate ring system isn't 
planar something may be wrong in the setting.

There's no easy solution - this is why it's considered an advanced topic. It 
is, however, very important. I've encountered a ligand that leaves its 
binding site during a simulation due to wrong parameters (in this case, the 
protonation of a protein side chain - FEBS  581, Pages 4120-4124, 2007).

Hope that helped,
Ran

On Fri, 27 Mar 2009 12:22:01 -0700 (PDT)
  "Josmar R. da Rocha" <bije_br at yahoo.com.br> wrote:
> Dear users,
> 
> I have been reading some posts about using externally computed charges to 
>replace Prodrg charges at ligand topology files. Many users commented on 
>the low trustability given to Prodrg charges (e.g 
>http://www.mail-archive.com/gmx-users@gromacs.org/msg02360.html ; 
>http://www.mail-archive.com/gmx-users@gromacs.org/msg17351.html ). Dr. 
>Verli pointed out the use of semi-empirical methods such as RM1 in cases 
>not involving simulations with sulphate or phosphate groups (what is not my 
>case) and the use of QM methods with the 6-31G** basis set, for example, to 
>obtain robust charges 
>(http://www.mail-archive.com/gmx-users@gromacs.org/msg03410.html). On the 
>other hand Dr. Mobley defined as a "a bad idea to compute charges for an 
>all-atom case using QM and then try to convert these to a united atom force 
>field". Other users advice that the best charges are that compatible with 
>the force field parametrization
> (http://www.mail-archive.com/gmx-users@gromacs.org/msg10760.html ; 
>http://www.mail-archive.com/gmx-users@gromacs.org/msg08308.html), usually 
>pointing to http://wiki.gromacs.org/index.php/Parameterization. Dr Friedman 
>suggested that "to calculate the electrostatic potential over the whole 
>molecule, and fit the atomic charges so that they reproduce this potential" 
>in order to make it less sensitive to small changes in the geometry of the 
>molecule may give good results 
>(http://www.mail-archive.com/gmx-users@gromacs.org/msg08308.html). Dr. 
>Lemkul stressed the need for charges refinement to reproduce 
>experimentally-observed behavior while trying to use QM charges with Gromos 
>ff. since "Parameterization under Gromos usually involves empirical 
>derivation of physical parameters, and free energy calculations using 
>thermodynamic integration". 
>Few examples of protein-ligand studies using Gromacs and Gromos96 ff that I 
>have access (from literature) seem to treat it as "take it for granted" 
>issue (any reference with a more detailed description would be welcome 
>:-)). Despite reading on this topic I could not compile all the information 
>in a clear and objective way (may be because I'm in the wrong track). Let 
>ask you some question that I find would help me to make my ideas more 
>clear:
> 
> 
> 1-am I overestimating the importance of ligand charges in such a simple 
>study of protein-small molecule (containg charged Phosphate groups) 
>complex? or
> 
> 1.1-The only way to test for this is doing many different simulation on 
>the same system using different type of computed charges to see what 
>happen?
> 
> 2-How could I try to choose a method to obtain reasonable charges based on 
>the reproduction of experimentally-observed behavior if I do not have 
>experimental data for my system?
> 
> 3-I also would like to know from users dealing with protein-ligand 
>interactions studies what do you consider a good approach to address this 
>problem?
> 
> Based on what I read I'd have a tendency to use HF/6-31G** ESP derived 
>charges (with necessary changes as to make it united-atom charges and 
>scaling that to a integer number for each group). Please, let me know if 
>that strategy would be as good as a disaster! 
> 
> Thank you very much for the attention.
> 
> 
> Josmar Rocha
> 
> 
> 
>      Veja quais são os assuntos do momento no Yahoo! +Buscados
> http://br.maisbuscados.yahoo.com