From: Courtney Taylor (courtney.b.taylor_at_gmail.com)
Date: Tue Sep 28 2010 - 15:02:34 CDT
Thanks for the links and quick response! I had found the "good practices"
publication a few days ago, but the wiki link is also helpful. I will see
how my tests go. And you are probably right about the window size, mine were
not very small for testing purposes.
While I'm thinking about it, to your knowledge, would selecting lambda
values based on Gaussian weighting be useful? I found it in the Amber
manual. It would remove the end-point catastrophe since lambda = 0 isn't
actually run. I probably couldn't use the namd2_ti.pl script for
post-analysis, but that shouldn't be an issue since dU/dlam*w = delG.
Î» w 0.01592 0.04064 0.08198 0.09032 0.19331 0.13031 0.33787 0.15617
0.5 0.16512 0.66213 0.15617 0.80669 0.13031 0.91802 0.09032 0.98408
Department of Chemical and Biomolecular Engineering
"Though the course may change sometimes the river always meets the
On Tue, Sep 28, 2010 at 2:51 PM, JÃ©rÃ´me HÃ©nin <jhenin_at_ifr88.cnrs-mrs.fr>wrote:
> Hi Courtney,
> You raise an interesting question, however there is no general
> agreement on the answer. You may be interested in discussions of "good
> and "best practices" (gosh, those guys keep raising the stakes):
> Consider Guideline 3 in the second document, which amounts to saying
> that alchElecLamdbaStart should be equal to alchVdwLambdaEnd. The idea
> is not backed by formal proof or published numerical data, and you'll
> note that the authors of that page were careful to call it a
> guideline. Yet, another argument in favor of this is that it makes the
> alchemical process conceptually simpler by identifying separate
> This choice is also coupled with complex decisions on how many
> lambda-point to use and how much to sample each of them. The errors
> you run into seem to be linked to an insufficient number of
> As you can see, there are so many parameters at play that so far it
> has been impossible to define an optimal recipe. Fortunately, for each
> problem, there is a range of reasonable choices leading to simulations
> that behave quite well.
> On 28 September 2010 17:50, Courtney Taylor <courtney.b.taylor_at_gmail.com>
> > I have a question regarding the appropriate selection of
> > alchElecLambdaStart. I am working on simulations of a protein in water
> > over a surface to calculate relative binding free energy using the
> > alchemical TI method (not FEP). I gradually mutate a residue from
> > (say, tyrosine) to mutant (alanine). I completed the tutorial that looked
> > this same mutation with solvation free energy and was able to reproduce
> > results.
> > I've run various test simulations, but run into issues when I test
> > the electrostatic lambda start. With decouple off, varying this is not an
> > issue. With decouple ON, however, if I lower this to 0.1 I get extremely
> > high âˆ†âˆ†G values (300-400 kcal/mol, when I expect ~10 or less). I also get
> > error messages from namd2_ti.pl stating that the integral for the window
> > too high.
> > So my question is: What is the logical way to select alchElecLamdbaStart?
> > there a specific reason why the default is 0.5? Is there a calculation I
> > perform to optimize this selection--i.e. in the user guide it states "As
> > lambda increases, once the particles are repelled it becomes safe to turn
> > TI electrostatics" so is there a way to estimate when these particles are
> > repelled?
> > Correspondingly is there a logical way to set alchVdwLambdaEnd other than
> > the default of 1.0?
> > Thanks. I want to make sure I have all of the variables correct to get an
> > accurate result.
> > Courtney Taylor
> > PhD Candidate
> > Department of Chemical and Biomolecular Engineering
> > Vanderbilt University
> > (225)-771-8554
> > (615)-343-3257
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:56:10 CST