AMBER Archive (2006)

Subject: Re: AMBER: Ti calculations in dmso solvent

From: David Mobley (dmobley_at_gmail.com)
Date: Thu Feb 23 2006 - 10:58:17 CST

Vitor,

If you are doing this with an annhilation procedure (where charge and van
der Waals interactions are turned off) as in the tutorial, there may be some
details of your calculation that you need to look at more carefully. In
particular,

(1) I have serious concerns I've mentioned previously on this list about the
ability of the scaling Amber uses for turning off van der Waals interactions
to produce converged results for these sorts of calculations. (I had trouble
getting reproducibility; essentially my results for the van der Waals
portion of my calculations could double from one trial to the next, even
with fairly long calculations (500 ps at each lambda value).
(2) It is not clear to me what the correct way to do these sorts of
calculations for charged ligands is. Usually to use PME you require that the
system be neutral; it seems difficult to maintain this if the ligand whose
charges are being turned off is charged. The problem isn't as apparent with
cutoff electrostatics, but it's not obvious to me that this means that there
will be no problem. Perhaps someone else can comment on this?

If there is a clear reaction path for your ion to come in and out of the
receptor, you might try using an umbrella sampling approach to compute the
PMF for pulling the ion out of the receptor and then compute the binding
free energy by integrating the PMF. In this way you get around both of the
potential problems I mentioned above, plus you avoid the having to calculate
the (extremely large) energy of turning off electrostatic interactions in
the receptor and solvent. IMO the largeness of these energies for charged
species means that the error in deltaG will be quite large. (Benoit Roux had
a recent paper using PMF's for binding free energies where he mentioned that
free energy of turning off the ligand charges in water was about 800
kcal/mol, because it was quite charged; the free energy in the site was
similar. The correct binding free energy was perhaps 8 kcal/mol, which would
obviously be swamped out by a 1% error (which is very small for these
calculations!) in the water or protein components of the free energy. He
concluded that for his ligand, the PMF approach was the only reasonable way
to go).

I can't speak to the solvent model issue, but unless you are spending a vast
amount of computer time on this problem (i.e. perhaps running 10 ns
simulations at 30 different lambda value) you may need to look carefully at
whether your TI approach is giving you the accuracy you need to even be able
to say anything about whether the solvent model is working well or not.
Proper error analysis (including the fact that your timeseries are
correlated) is key in this sort of thing.

David Mobley
Dill Group
UCSF

On 2/20/06, Vitor Manuel Sousa F?x <vfelix_at_dq.ua.pt> wrote:
>
> Dear Amber users
>
> We are interested to calculate the binding association
> constant between a macrocyclic receptor and a carboxylate
> anion with charges +5 and -2 respectively. The
> experimental data were obtained in dmso solution and I
> want to estimate the value of the constant in this medium
> using the thermodynamic integration methods. The best way
> to do that it is carry out two independent simulations as
> described at amber home page for the transformation
> toluene -> nothing in water. For the dimethyl sulfoxide
> we are using the flexible model reported by kollman in J.
> Phys. Chem 102,8070, 1998 and provided with amber.
> Unfortunately the values of the calculated constants are
> largely overestimated when compared with experimental
> ones. We would like to know if there is a more
> appropriated DMSO solvent model for free energy
> calculations?. .. The charges for anion and receptor were
> calculated separately using the resp methodology. However
> other charge models such as am1-bbc, other starting input
> geometries in Gaussian were used. The resp charges were
> also calculated for receptor and anion together but the
> result was worsted then that obtained with charges
> calculated independently.
>
> Thank you very much in advance for your help
>
> Cheers
>
> Vitor Felix
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu