[gmx-users] Re: Free energy with Gromacs.

[Justin A. Lemkul]

Eudes Fileti wrote:
> Hello Justin, 
> I am facing a very similar problem to that you experienced and described 
> in (http://www.gromacs.org/pipermail/gmx-users/2008-February/032429.html).
> 
> I throw this question in the GMX forum and Berk has kindly helped me. 
> But reading the forum I realized that you already could be solved the 
> problem so that maybe I could help more directly.
> 

Please keep Gromacs-related discussions on the list.  If you had followed the 
rest of the thread you reference, you will find that my calculations were 
falling victim to a bug in Gromacs-3.3.1.  If you are using a newer version, 
then my situation is not applicable, since the problem has been fixed (IIRC) as 
of Gromacs-3.3.3.

If you are using version 3.3.1, re-run your simulations with a newer version.

> I have tried to calculate the free energy of transfer from benzene to 
> ethanol
> for a polyhydroxylated (24 OH's). This system has 24 hydroxyl groups, 
> and in 
> ethanol, there should be more than 20 solute-solvent hydrogen bonds 
> being erased simultaneously (not to mention the possible intramolecular 
> HB's).
> 
> The Dg/dlambda plot, for both, benzene and ethanol shows a 
> very high and narrow peak near lambda=0. In the case of ethanol 
> is worse due to the solute-solvent hydrogen bonds. 
> 
> I performed two sets of simulations, one for sc-power=1 and another 
> for sc-power=2, using the following protocol:
> 
> 1) I made disappear the electrostatic interactions turning off the 
> charges (by 200ps), 
> 2) At the sequence I made disappear the LJ interactions (for more 200ps) 
> 3) Finally I performed a run of 0.5ns.
> Correct me i this procedure is inappropriate.
> 
> To start, Berk said me that the use of sc-power=2 never is recommended. Ok!
> Secondly, then he gave me a good tip that I was not taking into account:
> Disappear the electrostatic interactions using hard-core instead soft-core.
> I did this and actually work (only in part).
> 
> When I used softcore to desappear the electrostatic interations, the 
> value of dg/dlambda 
> for lambda = 0 was ~160000. Following the tip of Berk, wiht hardcore I 
> got ~2000!

Right, you should only need soft-core for the LJ component.

> However when I needed to use softcore again, now to remove the LJ 
> interactions, 
> the value returned back to ~160000.`
> 

I don't understand what you mean.  Is the total area under the curve 160000? 
That is ridiculously high.

I don't know that I have the expertise to help you much more.  I am not entirely 
familiar with your methodology.  What I have done in my own work is run 21 
independent simulations at lambda points between 0 and 1 (5 ns each after 
equilibration), and integrated the resulting curve.  None of my dV/dl points 
ever approached that magnitude, but I can't comment on your specific case, 
because I don't know what you're doing.

-Justin

> Could you gimme some insigths to solve this problem?
> Best
> eef
> 
> P.S.: You can solve your problem of polyphenolic compound?
> 
> _______________________________________
> Eudes Eterno Fileti
> Centro de Ciências Naturais e Humanas
> Universidade Federal do ABC
> Rua Santa Adélia, 166 - Bloco B, Sala 1048
> 09210-170  Santo André - SP Brasil
> +55.11.4437-8408
> skype: eefileti
> http://cromo.ufabc.edu.br/~fileti/

-- 
========================================

Justin A. Lemkul
Graduate Research Assistant
ICTAS Doctoral Scholar
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================