VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Thu May 08 2008 - 11:39:07 CDT

Jufang,
  To be honest, the differences you're seeing do concern me,
but these could conceivably be caused by a very peculiar structure
or atom selection. This is why I want to get a copy of the structure
file you're performing these tests on, so I can see what atoms you've
selected, what sample points are being generated, and whether the
variation you're seeing would be explained by statistical likelihood
of sample points falling within other atom radii, at critical points,
e.g. due to differences in rounding between the platforms.

The x86 (PC) platform typically uses 80-bit floating point precision
for operations within registers regardless how one writes the code,
so it will get slightly different results than the other two platforms
which will do their arithmetic in 32-bit precision (which is how the
code is actually written). These small differences in rounding and
floating point precision can result in big differences in results in
the case when one does true/false inclusion/exclusion tests on values
that lie close to a critical boundary, e.g. a cutoff radius in a
molecular dyanmics force computation, or the intersection between
VdW radii in the case of the SASA calculation.

Anyway, please send me the structure file you're testing with so
I can have a closer look.

Cheers,
  John Stone
  vmd_at_ks.uiuc.edu

On Thu, May 08, 2008 at 12:26:38PM -0400, Jufang Shan wrote:
> John,
>
> When I use 500, I get the same variations as without using the "-samples" flag.
> PC: 21.49
> ia64: 17.77
> irix: 18.39
>
> But the variations do decrease when I increase the number of samples.
> Do you think the following variations are normal?
> 1000:
> PC: 18.95
> ia64: 16.47
> irix: 17.77
>
> 5000
> PC: 17.62
> ia64: 18.22
> irix: 18.58
>
> Many thanks,
> Jufang
>
>
> On Thu, May 8, 2008 at 12:08 AM, John Stone <johns_at_ks.uiuc.edu> wrote:
> >
> > Jufang,
> > If you end up seeing noteworthy variations with 500 samples, then
> > I'd appreciate it if you can send me a copy of the structure in
> > question.
> >
> > Thanks,
> > John Stone
> >
> > On Thu, May 08, 2008 at 12:04:19AM -0400, Jufang Shan wrote:
> > > Hi John,
> > >
> > > I am using 1.8.6 and I am loading exactly the same single structure using
> > > the same script.
> > >
> > > I will use 500 to see whether I get the same results.
> > >
> > > Many thanks,
> > > Jufang
> > >
> > > On Wed, May 7, 2008 at 5:59 PM, John Stone <johns_at_ks.uiuc.edu> wrote:
> > >
> > > >
> > > > Hi,
> > > > Which version of VMD are you using?
> > > > You should definitely be using more than 50 samples, that is far
> > > > too low for accurate results, normally one would use 500 or more.
> > > >
> > > > Another question is whether you're using exactly the same structures
> > > > in your comparisons. Are you loading a single structure
> > > > (e.g. PDB) or are you loading a trajectory? Be sure you're comparing
> > > > identical frames etc.
> > > >
> > > > Cheers,
> > > > John Stone
> > > > vmd_at_ks.uiuc.edu
> > > >
> > > > On Wed, May 07, 2008 at 04:52:47PM -0400, Jufang Shan wrote:
> > > > > Hi,
> > > > >
> > > > > I have been using the same script to calculate sasa on different
> > > > machines
> > > > > (platforms) but got different results. I was wondering has anyone
> > > > encounter
> > > > > similar problems before.
> > > > > For some residues, the difference is bigger; for some, small.
> > > > >
> > > > > Here is the sasa-calculation script:
> > > > > set sel [atomselect top "protein or lipids"]
> > > > > set surf [atomselect top "segid A2 and resid 342"]
> > > > > set sasa [measure sasa 1.4 $sel -points pts -restrict $surf]
> > > > > #set sasa [measure sasa 1.4 $sel -points pts -restrict $surf -samples
> > > > 50]
> > > > >
> > > > > Here is the result if I turn off flag "-samples 50". x86_64 gives the
> > > > same
> > > > > result as ia64.
> > > > > PC: 21.49
> > > > > ia64: 17.77
> > > > > irix:18.39
> > > > >
> > > > > The result is still different when I change the probe to 1.0
> > > > > PC: 29.3
> > > > > ia64:24.5
> > > > > irix: 25.7
> > > > >
> > > > > The result is also different when I use flag "-samples 50"
> > > > > PC: 26.94
> > > > > ia64: 13.39
> > > > > iris:25.47
> > > > >
> > > > >
> > > > > Best,
> > > > > Jufang
> > > >
> > > > --
> > > > NIH Resource for Macromolecular Modeling and Bioinformatics
> > > > Beckman Institute for Advanced Science and Technology
> > > > University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> > > > Email: johns_at_ks.uiuc.edu Phone: 217-244-3349
> > > > WWW: http://www.ks.uiuc.edu/~johns/ <http://www.ks.uiuc.edu/%7Ejohns/>
> > > > Fax: 217-244-6078
> > > >
> >
> > --
> > NIH Resource for Macromolecular Modeling and Bioinformatics
> > Beckman Institute for Advanced Science and Technology
> > University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> > Email: johns_at_ks.uiuc.edu Phone: 217-244-3349
> > WWW: http://www.ks.uiuc.edu/~johns/ Fax: 217-244-6078 

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
Email: johns_at_ks.uiuc.edu                 Phone: 217-244-3349
  WWW: http://www.ks.uiuc.edu/~johns/      Fax: 217-244-6078