[gmx-users] FEP

paloureiro at biof.ufrj.br paloureiro at biof.ufrj.br
Thu Mar 23 20:53:39 CET 2006

Hi David,

I am not an expert in FEP, but I have used that method to try to answer the
following question: if there was a free energy difference between a 
system with
a Na+ ion inserted by the best-potential method or randomly.
I performed a series of MD runs (several lambdas) with annihilation of the ion
(with reaction field electrostatics) and, then, another series with 
of the ion, starting with the last frame of the previous series of simulations
(Please, see the attached file - fep.jpg)
The problem was motivated by the fact that a Na+ ion was "inserted" in 
a rather
dehydrated environment - and I questioned the physical/biological relevance of
Maybe the errors associated with the rection field method are cancelled 
as long
as one is using a delta(deltaG/deltaLambda).
But I agree with you that a PMF will be a better choice for the particular
system under discussion.



> I am rather nervous about doing FEP/TI for disappearing charged molecules,
> as it is not at all clear to me that it is possible do this correctly with
> current methods. Perhaps someone else may be able to comment more, but at
> least with long range electrostatics (PME), systems are required to be
> neutral. It IS possible to apply PME to a "charged" system, but I think this
> is typically done by spreading a uniform neutralizing charge throughout the
> box. It isn't clear that this is really correct for FEP/TI, I don't think,
> since in that case you will have a charged molecule overlapping with some of
> the uniform neutralizing charge, which could give strange energy artifacts.
> Perhaps things are better if you don't use PME electrostatics, but there may
> be boundary effects.
> I need to personally dig in to the literature on this a bit more, but I
> would suggest doing so yourself unless you're very sure that doing FEP/TI
> will give you something meaningful on this system.
> Also, if you *do* go that route, I would recommend against slow growth, as
> it tends to be hysteretic. It's probably better to run a bunch of separate
> simulations at different lambda values.
> And even further, it is probably *not* a good idea to turn off the charges
> and LJ interactions at the same time. It works better to turn off the
> electrostatics first (without using soft core, as this generally is fairly
> smooth with the charge), and then turn off the LJ interactions using soft
> core (recommend sc-alpha=0.5, as sc-alpha=1.5 is usually too large for LJ).
> However, if I were doing it, I would probably try to do it by computing the
> PMF for pulling the zinc ions (or one zinc ion, which is easier) away from
> the protein. You can compute the free energy difference from the PMF. The
> unfortunate thing about this is that it involves sampling a lot of stuff you
> don't care about, but at least it is clear the electrostatics is correct.
> Alternatively, dig into the literature and try and figure out if there is a
> correct way to do the electrostatics in this case. (And please let me know
> what you find out, if you do!)
> Oh, and one more thing: If DO compute the free energy by annhilating the
> zinc ions, you probably should use restraints to keep them in a particular
> region as you turn off the interactions, otherwise they will have to sample
> the entire simulation box in order for you to get converged results.
> Good luck.
> David

This message was sent using IMP, the Internet Messaging Program.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: fep.JPG
Type: image/pjpeg
Size: 30425 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060323/7ce546e7/attachment.bin>

More information about the gromacs.org_gmx-users mailing list