Re: [AMBER-Developers] [gohlke.uni-duesseldorf.de: Fwd: Re: some questions about current test failures]

From: Holger Gohlke <gohlke.uni-duesseldorf.de>
Date: Sun, 04 Mar 2012 12:09:13 +0100

I forward an email from Nadine regarding the pbsa / decomp issue. Would this be a starting point for further searching?


Holger


>>>>>


Dear Quin,

we re-checked the differences in the polar solvation energy. We observed differences in all mm_pbsa/Examples test runs, in which the solvation free energy is calculated with the PBSA module. For the hybrid model test and the decomposition analyses the differences in PB energy were especially large. According to our investigations these differences were introduced by changes made between 27.02.2012, 5:35 pm (8ad8c24ef9016553bcc3e9830bb3aac040877873) and 03.03.2012, 1:37 pm (23df75b9a36f3d1fceedc94be921f94776c96ffa). During this time we only made changes in the mm_pbsa Examples, which cannot not have caused the observed differences. From the git log files we could deduce that you made several changes on the 1st and 2nd of March in AmberTools/src/pbsa, among others in interface.c, pb_exmol.F90 and pb_read.F90. Between the 27th of February and 1st of March also the files AmberTools/src/pbsa membrane.F90 and AmberTools/src/pb_force.F90 were modified. Could these modifications be the cause for the differences we observe? Would you expect differences of the size > 1 kcal/mol for the hybrid model and individual residues in the decomposition analyses, because of changes made in the calculation procedure?
 
Best regards
Nadine

On 04.03.12, Holger Gohlke wrote:
> On 03.03.12, case wrote:
> > On Sat, Mar 03, 2012, Holger Gohlke wrote:
> > >
> > > with a new "pull" from origin/master, the problem due to sander in the
> > > decomposition of mm_pbsa.pl is gone. Thanks to whoever solved this.
> >
> > I'd like more information, please:
> >
> > 1. What is your latest commit from git?
>
> These were my last two actions with git today:
>
> commit 23df75b9a36f3d1fceedc94be921f94776c96ffa
> Merge: 25f18c2 84b2097
> Author: Holger Gohlke <gohlke.hhu.de>
> Date:   Sat Mar 3 13:37:41 2012 +0100
>     Merge branch 'master' of git.ambermd.org:amber
>
> commit 25f18c2ef382b739df260d75e11c1beee753f587
> Author: Holger Gohlke <gohlke.hhu.de>
> Date:   Sat Mar 3 13:35:52 2012 +0100
>     Updated example 11 of mm_pbsa.pl now using the latest iAPBS version available within APBS1.3.
>
> > 2. Can you define what you mean by "the problem due to sander in the
> > decomposition of mm_pbsa.pl is gone"? 
>
> We had observed a problem with the decomposition of internal(!) energies in $AMBERHOME/src/mm_pbsa/Examples/05_MMPBSA_Decomp_Residue. This is gone now. 
>
> > All of the test cases still seem to
> > fail, and the differences (e.g. in the "idecomp1" test directory)
> > show differences in the surface area terms, not in the polar parts.
>
> In $AMBERHOME/src/mm_pbsa/Examples/05_MMPBSA_Decomp_Residue, the differences arise from the PB polar part whereas the surface part from GB agrees.
>
> >
> > So: what is it that now works that didn't work before? 
>
> See above: internal energy decomp in 05_MMPBSA_Decomp_Residue works again.
>
> > Do we have any way
> > to know at all what change to sander we are talking about?
> >
> > And, how can we figure out what is a "good result" for the test cases? Is
> > there a way to recover the PB settings that were used for Amber11?
> >
> > Maybe if everyone could focus on just the idecomp1 test, that would help.
> >
> > What I see it that the current code gives identical answers for all the
> > EGB terms, but divergent answers for the ESURF and related terms. [This
> > is at odds with the behavior reported by Holger, so we need details from
> > him.] Most of the ESURF diffs are small (which is why I thought they
> > might of come from new radii), but at lines 511 and 518 there are enormous
> > diffs (thousands of kcal).
> >
> > If anyone has the time, you could start from the "olderDusseldorf" branch
> > that James put together. This has the code before the Feb. 3 changes from
> > Dusseldorf, plus the changes James made during February. I believe that the
> > head of this branch passes the idecomp1 (and other decomp) tests. We need
> > to cherry-pick one-by-one the Dusseldorf Feb. 3 changes, plus all the Irvine
> > changes, to try to locate the origin the test failure.
> >
>
>
> We will compare our current version against the olderDusseldorf branch, also looking at idecomp1. 
>
>
> Best regards
>
>
> Holger
>
>
> >
> > Let's try to make this a true regression test: find parameters for the current
> > code that will give the same numerical answers as with Amber11, [or be *sure*
> > we understand why the earlier answers were wrong.]
> >
> > ...thanks...dave
> >
> >
> > _______________________________________________
> > AMBER-Developers mailing list
> > AMBER-Developers.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber-developers
>
> --
> ++++++++++++++++++++++++++++++++++++++++++++++++++
> Dr. Holger Gohlke
> Professor fuer Pharmazeutische/Medizinische Chemie
>
> Heinrich-Heine-Universitaet Duesseldorf
> Institut fuer Pharm. und Med. Chemie
> Universitaetsstr. 1
> 40225 Duesseldorf
> Germany
>
> Tel.:  (+49) 211-81-13662; Fax: (+49) 211-81-13847
> Email: gohlke.uni-duesseldorf.de
> URL:   http://cpclab.uni-duesseldorf.de
> ++++++++++++++++++++++++++++++++++++++++++++++++++
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers

--
++++++++++++++++++++++++++++++++++++++++++++++++++
Dr. Holger Gohlke
Professor fuer Pharmazeutische/Medizinische Chemie
Heinrich-Heine-Universitaet Duesseldorf
Institut fuer Pharm. und Med. Chemie
Universitaetsstr. 1
40225 Duesseldorf
Germany
Tel.:  (+49) 211-81-13662; Fax: (+49) 211-81-13847
Email: gohlke.uni-duesseldorf.de
URL:   http://cpclab.uni-duesseldorf.de
++++++++++++++++++++++++++++++++++++++++++++++++++
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sun Mar 04 2012 - 03:30:02 PST
Custom Search