Re: [AMBER] general question

From: Jason Swails <>
Date: Fri, 27 May 2016 08:57:54 -0400

On Thu, May 26, 2016 at 2:33 PM, George Tzotzos <> wrote:

> Following a 100 ns MD trajectory of a homodimer (trajectory 1), I’ve
> calculated the free energy of binding of two monomeric subunits of a
> protein using (single trajectory approach) (deltaG = 30 kcal/mol).
> I run a second 100 ns MD simulation the same homodimer in complex with
> identical ligands in each subunit (trajectory 2). I calculated the free
> energy of binding of the two subunits in the presence of ligands.
> Subsequently, I used to strip the ligands and generate new
> topologies. The calculated free energy of binding of the two subunits is
> considerably different from that calculated from trajectory 1 (deltaG = 40
> kcal/mol).
> It can be argued that this difference is due to the trajectories sampling
> different conformational spaces. However, I’m wondering if the process of
> stripping the ligands using perturbs the trajectory thus
> giving a miscalculation of free energies.

​ only works on (and produces) topology files -- it does
nothing with coordinates. So there is no way it perturbs trajectories.
Whether or not it implements an identical force field compared to
generating the prmtops in LEaP is a fair question. In all of the tests
I've run (which were quite extensive when validating the approach), the
answer is yes. In a very small number of cases, you may get slightly
different improper torsion ordering for a very small number of impropers if
those terms have ambiguous sorting order (but that's because in those
cases, LEaP itself is not internally consistent, so it's impossible to
match that behavior exactly).

But as you pointed out, you are changing too many things to make any kind
of statement about the root cause of the differences. If you want to
evaluate whether or not has an effect on results,
you should run the *same* trajectory twice -- once with topologies
generated exclusively through LEaP and another time with topologies
generated by As long as everything else is identical, you
can definitively assess the effect is having.

A final comment I'll make is that single-trajectory MM/PBSA results depend
only on the Lennard-Jones and charge parameters on each atom (in addition
to the list of bonds, angles, and dihedrals which are used to determine
nonbonded exclusions). The actual bonded parameters have no effect, since
their contributions to binding are exactly 0. This makes an
even less likely culprit.


Jason M. Swails
AMBER mailing list
Received on Fri May 27 2016 - 06:00:07 PDT
Custom Search