Re: [AMBER-Developers] Accuracy of the LCPO approximation

From: Scott Le Grand <>
Date: Thu, 17 Nov 2011 14:19:28 -0800

Hey David, 1 ns of sampling barely begins to cover evaluating the accuracy
of this approximation. In contrast, in my 1993 JCC paper, I randomly
generated tens of thousands of uncorrelated conformations, then calculated
their SASA with my algorithm, an *exact* but expensive algorithm out of
Berkeley, and the Still approximation. Just like similar approaches like
Kazunori Toma's Residues in a Sphere potential look right for native
conformations, once you drift away from the native state, both of these
approximations lose meaning. In the MolSurf case you see an overall
correlation coefficient of ~0.3 which is essentially random (mine was 0.98
or so). In fact, Clark Still himself offered me $14K to come up with
derivatives for my method in order to replace his. I did so, but I'm not
satisfied with their accuracy so I never published or took the money.

So the question you have to ask yourself here is: do you feel lucky? Are
you comfortable with using a potential term that is chomping at the bit to
pull you away from the native state into the bizarro universe? I'd post
the incriminating diagram but I'm at SC11 right now with no access to said
paper. It's probably in your library though. Heck, you can probably
download it.


On Thu, Nov 17, 2011 at 10:26 AM, David A Case <>wrote:

> On Tue, Oct 04, 2011, Scott Le Grand wrote:
> > As someone who wrote a *really* fast SASA approximation 18 years ago
> > (basically Shrake and Rupley on steroids), here's my two cents.
> >
> > The Still et al. approximation at the heart of GBSA has a correlation
> > coefficient of ~0.3 with the actual SASA. On the bright side, this
> > approximation has a derivative.
> >
> > So if you're happy with some indeterminate analytic function of the
> number
> > of atoms surrounding a given atom as a surrogate for accurately
> calculating
> > the SASA, carry on... I'm not.
> >
> I'm moving this from the amber list to the amber-developers list. What
> Scott
> said didn't agree with what I remembered from a decade ago, so I asked Dan
> to look into this question, and his initial results are attached.
> Basically, although there are some real limitations with LCPO, a
> description of it as an "indterminate analytic function" with low
> correlation to the actual SASA isn't correct for this particular test.
> Correlation coefficients of 0.86 and 0.99 are found in the two cases
> looked at
> here.
> So: finding out more about what types of tests lead to the bad results
> would
> be useful, and it would also be nice to know about other approximations
> that
> we should consider.
> ....dac
AMBER-Developers mailing list
Received on Thu Nov 17 2011 - 14:30:02 PST
Custom Search