Yet another alternative is to forgo TI entirely and use a re-weighting
method such as MBAR. The formalism rigorously allows you to extrapolate to
any arbitrary lambda value regardless of whether you actually generated
samples in that ensemble or not. Of course, as Jason pointed out, the
shorter the "distance" in lambda space that you extrapolate the more
accurate it will be, to the point that eventually the correction is well
below statistical noise.
The biggest drawback to this approach is the added cost of calculating the
energy of the samples at all of the other lambda values (which is not part
of TI). Fortunately, Thomas implemented this type of calculation on the
fly, see the "ifmbar" and related flags if you are interested (section
4.1.3 of the AMBER12 manual).
Regards,
Brian
On Mon, Sep 17, 2012 at 11:28 PM, Jason Swails <jason.swails.gmail.com>wrote:
> On Mon, Sep 17, 2012 at 7:25 PM, <psu4.uic.edu> wrote:
>
> > Dear Professor Case,
> >
> > This is Henry from Dr. Michael Johnson's lab. In Amber 11/12 user
> > manual, there is one sentence saying "When using softcore potentials, λ
> > values should be picked so that 0.01 < clambda < 0.99". I am quite
> > confused why there is such a setting. Could you kindly offer an
> explanation
> > or recommended papers for this?
> >
>
> This is not a suggestion for TI in general, but rather a suggestion
> specifically for the Amber implementation of TI in sander. In order to
> simplify coding, the mechanism that is used is that 2 separate topology
> files are used and a separate force calculation is carried out with both
> topology files (one for each end state). The forces are then scaled by
> 1/lambda and 1/(1-lambda) for the two end states so they can simply be
> summed up to give the net force on each atom.
>
> This has the advantage that it is very easy to code, but has the
> disadvantage that the scaling factors go to infinity at the two end-points
> (0 and 1). Therefore, for numerical stability of the Amber implementation
> of TI, you need to select lambda values between 0.99 and 0.01. However,
> both of these values are very close to their respective end-points, and I
> would expect that the effect of neglecting the last 0.01 of the alchemical
> coordinate in each direction is well below the noise of the method itself.
>
>
> >
> > One interesting thing is I found my previous error (
> > http://archive.ambermd.org/201208/0390.html), in which the mask atoms
> > becomes "a spider monster from space" , is due to clambda < and clambda
> > > 0.99. Alternatively, the softcore potential production run is quite
> > normal if 0.01 < clambda < 0.99, but not clambda < 0.01 and clambda >
> 0.99.
> >
> > If 0.01 < clambda < 0.99 in softcore potential is true, then is there
> any
> > recommended way to set the largest and smallest clamda values for a 12
> > windows TI run and following Gaussian integration? (The corresponding
> table
> > is in Amber 12 user manual's "Table 4.1.: Abscissas and weights for
> > Gaussian integration<
> > http://i1076.photobucket.com/albums/w454/happypsu4/table41.png>.")
> >
>
> Sure, just use 0.99 and 0.01 and call them 1 and 0 :). You also don't have
> to use Gaussian integration. There are a number of packages that will do
> numerical integration for you (I typically use xmgrace) if you don't want
> to implement your own integration scheme (e.g., trapezoid rule).
>
> HTH,
> Jason
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Candidate
> 352-392-4032
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
================================ Current Address =======================
Brian Radak : BioMaPS
Institute for Quantitative Biology
PhD candidate - York Research Group : Rutgers, The State
University of New Jersey
University of Minnesota - Twin Cities : Center for Integrative
Proteomics Room 308
Graduate Program in Chemical Physics : 174 Frelinghuysen Road,
Department of Chemistry : Piscataway, NJ
08854-8066
radak004.umn.edu :
radakb.biomaps.rutgers.edu
====================================================================
Sorry for the multiple e-mail addresses, just use the institute appropriate
address.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 18 2012 - 06:30:05 PDT
