Re: [AMBER] Total interaction energy between monomeric units of a dimer

From: Jason Swails <>
Date: Thu, 20 Aug 2015 09:12:14 -0400

On Thu, 2015-08-20 at 09:09 -0300, Hector A. Baldoni wrote:
> Hi Anu,
> Attached you will find some in/out files about how to calculate the
> interaction energies between protein-K+ using anal. You could download
> anal software, compile and learn from the attached file how to use it.
> I hope this will help you.

But this does not address Dave's points about the lack of pairwise
decomposability of the GB and PME potential energy functions.

Sure, anal will give you something it *calls* an interaction energy, but
that interaction energy will be a simple, gas-phase (maybe with an
aphysical, distance-dependent dielectric) interaction that bears little
resemblance to the potential energy function you actually run
simulations with these days.

In particular, the interaction between surface atoms of the two species
is treated exactly the same as the interaction between two buried atoms
near the interface. In a solvated environment, this is clearly the
wrong thing to do if what you are interested in is the interaction
energy in solution.

The challenge here is that the GB and PME potentials -- which account
for solvation effects in a more physically sensible model -- are not
pairwise decomposable, so the anal approach will not work here. You
need something different.

For GB, the "pairwise decomposition" coded in sander treats the
effective GB radii as a "constant" derived from the calculated effective
radii of the entire system (when in reality the radii differ between the
"bound" and "unbound" complexes). This approximation makes the GB
potential pairwise decomposable. While approximate, it is almost
certainly better than what anal does.

For PME, the reciprocal sum is definitely not pairwise decomposable, and
pair interaction energies need to be calculated by performing 3 separate
nonbonded energy calculations -- one with the full system, one in which
only the first part of the pair has nonzero nonbonded terms, and one in
which only the second part of the pair has nonzero nonbonded terms. You
still get contributions from the periodic images, so it's still an
approximation, but it's significantly better than what anal did.

I'd not suggest using anal for this purpose. The idecomp=3/4 option in
sander and the 3-calculation PME approach are substantially better, and
may give results that are even qualitatively different from what anal


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Thu Aug 20 2015 - 06:30:03 PDT
Custom Search