Re: FEP with flattened free energy surface?

From: Ivan Vyalov (VyalovIvan_at_yandex.ru)
Date: Fri Apr 08 2011 - 15:48:09 CDT

Well, I took a protein as a common example, but the system is actually cellobiose in ionic liquid. That's why I have these differences and values of free energy don't change with time, so I think it's ok with respect to converrgence. But if I apply bias calculated at the point where the barriers are high, then in point where solute-solvent interactions are switched off my system will be over-biased, won't it?

09.04.11, 00:05, "Jérôme Hénin" <jhenin_at_ifr88.cnrs-mrs.fr>:

> Hi Ivan,
>
> On 8 April 2011 21:12, Ivan Vyalov <VyalovIvan_at_yandex.ru> wrote:
> > Hi Jerome,
> >
> > Thanks for reply! So, colvars config should contain lines like these:
> >
> > abf {
> >  colvars phi psi
> >  updateBias no
> >  inputPrefix colvarA01o
> >  fullSamples 200
> > }
>
> This looks fine. Setting the fullSamples parameter is a good idea, actually.
>
> > And then I start FEP calculations that seems to be working ( I didn't try with solvent yet ).
> >
> >
> > May I ask you one more question?
> >
> > If I want to perform annihilation of dipeptide then won't I have to recalculate bias ( along \phi & \psi ) between FEP steps ( perhaps, not between every lambda window )? In my case solvent changes free energy surface quite significantly and if I have barriers ~6 kcal/mol for vacuum then for solvent they are ~16 kcal/mol.
>
> Well, you probably don't need the biases to be perfectly accurate,
> they are just there to improve sampling. It is a bit surprising that
> solvent effects on conformational free energy are so drastic. Are you
> confident that the ABF calculations are converged (or close enough)?
>
> In principle you can certainly apply different biases at different
> stages, if you are ready to run more calculations. I am not sure it is
> worth the effort. It depends on the specifics of your system (and how
> much time you are willing to spend on it).
>
> By the way, an idea just came to my mind: it might be a better idea to
> replace the data in the gradient file with that in the '.est' file
> written by abf_integrate. The good thing about that data is that it is
> a "true gradient" (its curl is zero), whereas the raw data in the
> '.grad' file is not, due to statistical noise. Applying the bias based
> on the '.est' file ensures that the biasing force is conservative.
>
> Best,
> Jerome
>
>
> > 06.04.11, 20:45, "Jérôme Hénin" <jhenin_at_ifr88.cnrs-mrs.fr>:
> >
> >> Hi Ivan,
> >>
> >>  That should be possible. Here is how I see it:
> >>  1) run ABF to get biasing potential
> >>  2) run FEP with ABF bias, disabling updatePotential to keep the bias
> >>  time-independent.
> >>  In this step, save FEP data and colvars trajectory with the same
> >>  frequency (e.g. 10 timesteps) to make the next step easier
> >>  3) compute a FEP estimator with unbiased data: you need to reweight
> >>  the ensemble averages based on the applied ABF bias.
> >>
> >>  Cheers,
> >>  Jerome
> >>
> >>  On 6 April 2011 11:49, Ivan Vyalov <VyalovIvan_at_yandex.ru> wrote:
> >>  > Hi all,
> >>  >
> >>  > I wonder, is it possible to perform FEP calculations with solute molecule with flattened free energy surface along several reaction coordinates?
> >>  >
> >>  > For instance, if I take dipeptide with conformational degrees of freedom ( dihedrals \phi & \psi ) and two minima defined by them, I should take into account this minima and do something similiar to what is described in JPhysChem 97, 3409.
> >>  >
> >>  > From the other hand I could run ABF, apply forces and perform FEP with flat free energy surface along these coordinates, is it possible in NAMD? Or I should do such task in a different way?
> >>  >
> >>  >
> >>  >
> >>  > --
> >>  > with best regards,
> >>  > Ivan
> >>  >
> >>  >
> >>
> >>
> >>
> >
> > --
> > with best regards,
> > Ivan
> >
> >
>
>
>

-- 
with best regards,
Ivan

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:56:57 CST