[gmx-users] g_sas: unable to reproduce data from original article of the DCLM (Eisenhaber1995)

João M. Damas jmdamas at itqb.unl.pt
Tue Mar 4 04:39:50 CET 2014


Hello Teemu,

Thank you! I am glad the GROMACS team is concerned about this particular
problem.
And I am sorry for my late reply.

For the sake of conciseness, so we do not get lost in the data, I am only
showing the single precision results (we may bring double precision later
on again). Also, since GROMACS 4.0.4 is flawed in the -pbc flag, as Teemu
said, I also excluded it from these results (it is not going to get
corrected anyway).

I compiled a GROMACS version 4.6.3 with the fix presented in
https://gerrit.gromacs.org/#/c/3199/ (in single precision).

I repeated the tests and the results are presented in Attached Table 3.

   - the fix brings the -nopbc results close (within 1 square angstrom) to
   the ones obtained by ASC, except for 1TIM where the difference is within
   10-20 square angstrom. Maybe 1TIM has a difficult shape, because ASC also
   had the largest difference between numerical DCLM and the analytical method
   in this protein.
   - the "-pbc yes" results though, do not change (which is normal,
   considering the flag is working in this version and the fix is inside a
   conditional expression that depends on the -pbc flag). It continues to
   baffle me why this is happening. If the molecule is inside the box and far
   from the edges (again, a test using a .gro created by editconf -d 5 gave
   the same exact table as Attached Table 3), I see no reason why these
   results should be different...

So, I think Bug #1445 <http://redmine.gromacs.org/issues/1445> is indeed
solved, but we still need to solve an unknown bug with "-pbc yes". Should I
open a new bug related to this in the redmine?

I can perform more tests, if anyone has more ideas.

Best,
João

--
Attached Table 3
met  - method used, anl=analytical method, num=numerical method (DCLM)
dots - number of points used for the numerical method
src  - source of the data, paper=Eisenhaber1995, ASC=ASC software,
       g463{sp,fx}=g_sas of GMX4.6.3 {single precision,fix
3199<https://gerrit.gromacs.org/#/c/3199/>
}
pbc  - PBC flag for g_sas, yes=-pbc, no=-nopbc
Areas are in square angstrom
======================================================
  met  dots     src  pbc      4PTI      3FXN      1TIM
======================================================
  anl         paper        3973.80   6943.80  20002.80
  anl           ASC        3973.81   6943.80  20002.80

  num   122   paper        3961.40   6968.30  19970.90
  num   122     ASC        3961.44   6968.33  19970.90
  num   122  g463sp  yes   3825.86   6257.83  19867.90
  num   122  g463sp   no   4169.21   7731.63  24080.20
  num   122  g463fx  yes   3825.86   6257.83  19867.90
  num   122  g463fx   no   3957.81   6955.99  20004.20

  num   362   paper        3971.80   6933.40  19997.10
  num   362     ASC        3971.79   6933.37  19997.10
  num   362  g463sp  yes   3838.35   6248.43  19886.80
  num   362  g463sp   no   4195.02   7704.70  24132.10
  num   362  g463fx  yes   3838.35   6248.43  19886.80
  num   362  g463fx   no   3969.92   6936.20  20016.70

  num   642   paper        3967.90   6944.40  19998.70
  num   642     ASC        3967.78   6944.37  19998.70
  num   642  g463sp  yes   3831.64   6255.64  19882.70
  num   642  g463sp   no   4188.63   7701.99  24119.40
  num   642  g463fx  yes   3831.64   6255.64  19882.70
  num   642  g463fx   no   3966.21   6941.26  20013.60

  num  1002   paper        3974.10   6939.10  20012.20
  num  1002     ASC        3974.07   6939.12  20012.20
  num  1002  g463sp  yes   3840.58   6250.14  19897.40
  num  1002  g463sp   no   4193.60   7700.64  24143.10
  num  1002  g463fx  yes   3840.58   6250.14  19897.40
  num  1002  g463fx   no   3973.51   6939.68  20027.50

  num  1472   paper        3975.70   6943.30  19997.00
  num  1472     ASC        3975.70   6943.35  19996.90
  num  1472  g463sp  yes   3841.69   6256.92  19882.80
  num  1472  g463sp   no   4198.08   7709.30  24123.30
  num  1472  g463fx  yes   3841.69   6256.92  19882.80
  num  1472  g463fx   no   3975.88   6943.96  20012.40
======================================================




On Fri, Feb 28, 2014 at 4:49 AM, Teemu Murtola <teemu.murtola at gmail.com>wrote:

> Hi,
>
> On Thu, Feb 27, 2014 at 4:09 PM, Teemu Murtola <teemu.murtola at gmail.com
> >wrote:
>
> > I'll post additional information once I have had time to test more cases.
> >
>
> I have now created <http://redmine.gromacs.org/issues/1445> to track the
> issue, and uploaded a fix to Gerrit: <https://gerrit.gromacs.org/#/c/3199/
> >.
> My current set of tests is here: <https://gerrit.gromacs.org/#/c/3200/>
>
> Looking at the history, as described in the Redmine issue, this should
> explain incorrect results with -nopbc. The Redmine issue also provides an
> explanation for why 4.0.4 gives the same results with -pbc and -nopbc. This
> doesn't explain why it gives a slight underestimation with -pbc in 4.6.3,
> but you could try repeating the tests with the fixed version to see whether
> the PBC treatment or something else is giving rise to the underestimation.
>
> Best regards,
> Teemu
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>



-- 
João M. Damas
PhD Student
Protein Modelling Group
ITQB-UNL, Oeiras, Portugal
Tel:+351-214469613


More information about the gromacs.org_gmx-users mailing list