Re: FEP minimum requirements

From: Brian Radak (
Date: Thu Mar 01 2018 - 08:18:27 CST

There is not, to my knowledge, any positive test for convergence for any
kind of MD simulation, much less FEP. The forward/backward test performed
by ParseFEP can only serve as a negative test for convergence - if they
don't match, then one or both of the simulations are not converged.

In my personal experience, which doesn't really cover absolute ligand
binding free energies, I would always prefer running longer per window than
running more windows.

Running a bidirectional simulation and using SOS or BAR is almost certainly
better than one sided FEP and should give more realistic uncertainties.


On Thu, Mar 1, 2018 at 4:05 AM, Francesco Pietra <>

> Hi Brian:
> I went on with that FEP of previous thread "Ligand atoms moving too fast
> with FEP" protein-ligand complex. Aim: estimating the binding free energy
> with respect to other ligands for the same binding site.
> frwd Bound was run with delta lambda 0.2, i.e., five windows, with 100,000
> steps re-equilibration and 400,000 steps FEP at each window, ts=1.0fs.
> Is it possible guessing from parseFEP, from simply frwd, if such
> conditions lead to acceptable convergence? Or back running is also needed
> for such a guess? Or, finally, should be better running frwd, or
> forwd+back, with ten or more windows?
> From the attached summary.png, from parseFEP, for the above frwd, I fear
> that the above conditions for frwd are inadequate (convergence far from
> being attained) to my purpose. All such questions because I am dealing with
> a large system, with costly simulations.
> Thanks a lot for advice.
> francesco pietra

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2019 - 23:19:43 CST