From: Floris Buelens (floris_buelens_at_yahoo.com)
Date: Tue Mar 31 2009 - 04:29:56 CDT
> But, this is a trick, so I want to do the thermodynamic integration as well.
Just to emphasise, FEP and TI are two ways of accessing a single quantity, the free energy change associated with an alchemical transition. The free energy change is the only physically meaningful result, and TI doesn't provide any additional insights as such.
> I would assume that the lack of charges for some values of lambda is
> somehow taken into account in the thermodynamic integration, but I
> would feel better if I knew how.
The delayed scaling of charges is handled transparently by both FEP and TI. When lambda <= elecLambdaStart, electrostatic interactions of the lambda-coupled atoms are switched off (multiplied by zero). With FEP, the free energy difference to the next 'window' will only include an electrostatic component if lambda > elecLambdaStart in that next window. With TI, the derivative of the electrostatic potential (dU/dlambda) is calculated even when lambda <= elecLambdaStart, but the integration should be performed only for elecLambdaStart < lambda < 1 (this is handled automatically by the integration script, NAMD_ti.pl).
> How would you know if you were using an inappropriate value of
The main danger that has to be avoided is having exposed charges without a repulsive core (e.g. at least a partial van der Waals interaction). If this happens, you might end up with a situation where particles of opposite charge overlap and become trapped in an 'infinite' energy well. This eventuality should be covered by all but very small values of elecLambdaStart. If in doubt, the default values of elecLambdaStart and vdwLambdaEnd (0.5 and 0.5) are always risk-free.
----- Original Message ----
From: "brmorgan_at_clarku.edu" <brmorgan_at_clarku.edu>
To: Chris Harrison <char_at_ks.uiuc.edu>
Sent: Monday, 30 March, 2009 19:21:34
Subject: Re: namd-l: use of tiElecLambdaStart (NAMD 2.7b1)
Yes, I would have preferred to do a perturbation which preserves
electrostatic neutrality, but it did seem much more complicated. It is a
potential issue in my case, since the correction to the free energy (the
Wigner self energy term) is 3x greater than the final delta G given by
NAMD 2.6's FEP calculations. However, the mutations I make involve an even
number of charges, so I am running the FEP again, this time with enough
counter ions to make the charge difference symmetric about zero (+2 ->
-2), which would make the correction term vanish (theoretically). But,
this is a trick, so I want to do the thermodynamic integration as well.
I am concerned regarding tiElecLambdaStart because the change in energy in
my system is dominated by the electrostatic interactions. I would assume
that the lack of charges for some values of lambda is somehow taken into
account in the thermodynamic integration, but I would feel better if I
knew how. Also, in the sample input file in the user guide,
tiElecLambdaStart is set to 0.1, but the default is 0.5? How would you
know if you were using an inappropriate value of tiElecLambdaStart? Not
having naked charges as Jerome explained sounds like a good idea, so I do
think that I should use it, but I really want to understand what is going
on from a theoretical perspective.
> After re-reading my earlier email I think my wording may cause
> unnecessary concern from you. While analytically it is desirable to
> a neutral charge in PME, the implementation in NAMD effectively
> neutralizes forces resulting from the "infinite propagation" of charge.
We can discuss how in more detail if necessary, but this does permit for
example such perturbations as Leu -> Lys within a given simulation run.
> probably should have said "While not balancing the charge may yield
reasonable or correct results, the analytical propogation of an infinite
charge in PME is generally felt to be bad form." The key being the
difference between an "analytical" treatment and as treated in the
implementation in NAMD. If you can easily do a perturbation and
> net neutral charge without introducing additional complexity or degrees
> freedom, then it's probably a good idea to do it; however in biological
systems it is often debatable if this can be done easily without
> Chris Harrison, Ph.D.
> Theoretical and Computational Biophysics Group
> NIH Resource for Macromolecular Modeling and Bioinformatics
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:21:51 CST