From: Anthony Cruz-Balberdy <>
Date: Sat, 09 Nov 2013 18:57:25 -0500


Sent from my Galaxy S®III

-------- Original message --------
From: David A Case <>
Date: 11/08/2013 12:16 PM (GMT-05:00)
To: AMBER Mailing List <>
Subject: Re: [AMBER]
On Fri, Nov 08, 2013, Anthony Cruz Balberdi wrote:

> but there is a huge difference with the POISSON BOLTZMANN
> part (my results:DELTA TOTAL -29.6742(AVE) 7.9567(STDD)
> 1.1252(STDErr);tutorial: DELTA G binding -86.3878(AVE) 8.1828(STDD)
> 1.1572(STDErr)).

What has happened is that the PBSA codes in Amber changed the default
settings during one of the releases.  (The changes were in how the
nonpolar contribution to the interaction with implicit solvent is
estimated, not to the electrostatic component.)  This has caused no end of
problems, since running old input files (e.g. from the tutorials) with new
codes now gives a different answer.  Three points about this:

1. To reproduce the tutorial, you will have to figure out the settings that
were used in the original run, and explicitly set them in your inputs.  (There
may be a bug fix in the works that will make this easier. Volunteers are also
solicited to update the tutorials.)

2. There is always some tension between maintaining backward
compatibility, and trying to nudge users towards using more modern (and
often "better") parameters and protocols.  One way to straddle this fence
is to provide no default (where there used to be one) so that current
users are required to think about it and make a choice.  We didn't do
this for pbsa, and instead just changed the default behavior, causing
much confusion.  Users can hope that Amber developers have learned their

3. The fact that the old and new results (even for the *binding energy*) are
so wildly different (by 50 kcal/mol!!) should send you running and screaming
away from using mm-pbsa altogether, or at least to think hard about what is
going on.  If I can quote from a recent review article that I wrote
(Biopolymers 99:969-977, 2013):

"Despite many promising results, the general performance of the MM-PBSA model
in studying ligand binding to proteins and nucleic acids has not been good
enough for straightforward use as a part of a ligand design strategy. Beyond
the obvious (but persistent) difficulty in obtaining good binding poses in the
absence of experiment, a part of this probably arises from the large
cancellation noted above between direct electrostatic effects and solvent
screening; unless this balance is accurate (for many different conformations)
trends in binding energies involving only a few kilocalories per mole may be
lost. Continuum solvent models almost certainly fail as well to capture
localized hydration effects or trends in small cavities where the size of an
individual water molecule is key to understanding what “fits” and what does
not. Beyond this, the need to estimate changes in configurational entropy is a
weakness, as there is no general procedure for making such calculations."

The article (briefly) discusses some other good and bad points of implicit
solvent models and MM-PBSA calculations, and points to other review articles.
I don't want to start a flame-war here, but Amber users should understand that
(a) MM-PBSA is an advanced method, and you need to both experience and
patience to apply it well; (b) even absent any technical glitches, YMMV
concerning the physical realism of the resulting free energy estimates.

...dac  (mainly my $.02)

AMBER mailing list
AMBER mailing list
Received on Sat Nov 09 2013 - 16:00:02 PST
Custom Search