From: Felipe Merino (felipe.merino_at_mpi-muenster.mpg.de)
Date: Mon Jan 05 2015 - 10:27:30 CST
As far as I can tell your problem comes from how dG is actually
calculated. dG involves an exponential average of dE. Since dE changes
from 14 to -9 it means that one of your data points "weights" e**23
times more than the other and of course it'd take quite some sampling to
bring it back to normal.
Ideally, you'd want that the variance of the dE inside your window stays
in the order of kT, otherwise you end up calculating your dG from the
tails of the distribution which are exactly the configurations you did
not sample properly.
On 05/01/15 16:44, Jonasson Gabriella wrote:
> Dear NAMD users,
> I'm trying to estimate the free energy of binding of a small molecule to a protein by means of FEP simulations. I'm using NAMD 2.9 and the parameters proposed in the tutorial "Protein:ligand standard binding free energies" by Gumbart, Roux and Chipot.
> I have problems with the dG sampling in some of my simulation windows. Sometimes I see a huge change in dE from one step to the next (e.g. +14 -> -9 kcal/mol). I can't find the structural origin of this energy change (some sort of clash maybe?) because the coordinates aren't printed out as often as the FEP-energies. Anyhow, the energy difference between the two states goes back to "normal" (i.e. ~14 kcal/mol) pretty quickly (~400 fs) and this episode doesn't seem to affect dE_avg much at all. On the other hand, dG is completely put off track, where it stays for the rest of the simulation. I’ve attached the graphs that illustrate this.
> I know that dG is based on an ensemble average but I don’t understand why it doesn’t behave more similarly to dE_avg under these conditions. Can anybody explain?
> Is this kind of simulation useless or can I post-treat the energies somehow?
> Best regards,
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:20:47 CST