AMBER Archive (2006)

Subject: Re: AMBER: Question about free energy with mm_pbsa

From: Thomas Cheatham (
Date: Fri Sep 01 2006 - 01:23:59 CDT

> Question 1: While we use mm-pbsa for calculating the delta G, both GB and
> PB do not minimize the complex, receptor and ligand. The complex snapshots
> are at a relatively low energy conformations, but those snapshots of

As dac mentioned, this is a feature; stated another way, if you run
separate trajectories for each, differential relaxation in each
(potentially unrelated to the binding) may lead to more noise. For
example, maybe in the simulation of the complex, some motion in the tail
unrelated to the binding leads to effectively a lower energy state. If
this same motion is not sampled in the "free" receptor (in a multiple
separate trajectory approach), this will tend to overstabilize the
binding. Even with single trajectory approaches, if you are trying to
compare multiple different ligands and run a series of different
ligand-bound complexes, each may differentially "relax". This is
discussed a little bit in Spackova et al. JACS (2003) in the supplementary
material and in other MM-PBSA papers and reviews.

> Question 2: In mm-pbsa NM section, if we use the same number of snapshots
> as we used for delta E calculation, it will take too long time to finish the
> Nmode calculation. For the continuous snapshots, after minimization, I can
> imagine that many snapshots would share the same conformation. So is that
> okay if we use fewer snapshots for Nmode calculation, let's say 1 in about
> 10 snapshots?

Clearly the cost of NMODE calculations on each frame is prohibitive. If
you look at the early MM-PBSA papers and even more recent, a much smaller
set of frames are investigated by NMODE, or quasi-harmonic estimates of
the entropy are applied (for example papers of Laughton and/or Orozco).
If I were forced to judge the question of how many are necessary, I would
have to do some statistical analysis and maybe block average comparisons
for my particular case.

> Question 3: After the simulation system reached equilibrium, how many
> snapshots we should use for delta G calculation? What is the reasonable
> period?

I do not intend to sound pessimistic and I will not claim I have all the
answers here, but the best thing to do is to test and find out. Also, I
would careful with assumptions that you have reached equilibrium, and more
importantly, how you judge if you have attained equilibrium. Some
properties converge very rapidly, others will not converge in 10-100ns

If you look to the literature, "reasonable periods" in common usage range
from 100 ps in some papers to 10's of ns in others. Longer does not
necessarilly mean better as differential relaxation in the stochastic MD
trajectories of various complexes can lead to noise or artifactual
calculated differences in the estimated energetics.

Remember that these energetic estimates are only one tool in the arsenal
for attempting to understand what is happening in a simulation. Also note
that in addition to the MM-PBSA papers by many groups, consider looking at
related methods that look at averages over trajectories, such as the ES/IS
methods of Hermans, linear interaction energy approaches of Aqvist, etc.


The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to