AMBER Archive (2006)

Subject: Re: AMBER: error estimate for free energy perturbation calculation

From: Chris Moth (
Date: Wed Oct 25 2006 - 11:58:14 CDT wrote:

> Dear all,
> Can someone please recommend references for the estimate of error
> (including sampling error) for the free energy perturbation calculation?
> I want to know how reliable my delt_delt_G is, and how many trajectories
> are necessary, how small each step should be, etc.
> Thanks so much!
> Jeny

Ths is a question of great question to me, as well. I hope you get a lot
of answers.

There is some discussion of this in Leach's, "Molecular Modelling
Principles and Applications" Section 11.6 "Potential Pitfalls with Free
Energy Calculations", along with important caveats.

This recent article was interesting to me: "Free energy perturbation
approach to the critical assessment of selective cyclooxygenase-2
Journal of Computer-Aided Molecular Design 19: 1731, 2005

and it states

"Starting from the equilibrated structures obtained from the
aforementioned procedure, we performed the production dynamics of
for COX-2 in complex with the inhibitors shown in Figure 3. Each
perturbation consisted of
21 windows with 5000 steps of equilibration and 10,000 steps of data
collection. ..... The free energies reported in the remainder of this
paper always refer to the forward direction, i.e., in the direction from
k 0 to k 1. As an estimation of the statistical error, we used half
the difference between the absolute values of
forward and reverse free energy changes, and the error in DDGbind is
calculated as the square root of the sum of the squares of the
individual errors in
DGunbound and DGbound. "

In my own experimentation with FEP, I found I could get acceptably small
error increases (~.1kCal or so) with larger lambda steps in the middle
points of the trajectory. delta-Lambda need not be constant. I suppose a
lot could depend on one's particular system.

I found it interesting that the above authors got dGdG binding errors
less than .8 kCal/mol - but I am also thinking that perhaps their
equilibration and data gathering runs might have been a bit on the short
side (see Leach).


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