Re: [Cdk-devel] ModelBuilder3d/aromaticity detection

[Stefan K. <ste...@eb...>]

On Sunday 07 September 2008 21:50:18 you wrote:
> On Fri, Sep 5, 2008 at 12:26 AM, Stefan Kuhn <ste...@eb...> wrote:
> > Dear co-developers,
> > two comments which might be important for the upcoming release:
> > 1. I fixed the ModelBuilder3d - all tests except one work (will look at
> > this tomorrow). The reason were changes in the fingerprinter,
>
> Such as? I think Christoph regenerated fingerprints too, after which
> the main FingerPrinter did no longer change... how come the index is
> again broken then? (It's been puzzling me...)
Don't know. Just found that regenating the fingerprints helps (with the proper 
settings for the TemplateExtrator, which is what i missed last time I tried, 
so it didn't help then).
>
> > apart from the
> > ill-fated changes in the modelbuilder itself, which I fixed earlier. I
> > can only ask everybody to run tests (I know I might have missed that
> > myself). I added a test case to the TemplateHandler3DTest, which tests
> > the fingerprints for stability.
>
> Please provide some information in what was/is causing fails...
As I said, some change in the fingerprinter (might be aromaticity stuff, but I 
don't know).
>
> > Worse, when running the ModelBuilder3DTests with a slight change, I
> > get errors "org.openscience.cdk.exception.NoSuchAtomTypeException: The
> > AtomType Csp2 could not be found". It looks like the aromaticity
> > detection (now?) relies on atom typing,
>
> Sure... one needs to know if all atoms in the possibly aromatic ring
> are sp2 or planar3 (N, O) and how many pi electrons they contribute to
> the ring system.
>
> > but atom typing fails on aromatic carbons.
>
> In what way? With the above Csp2 atom type missing? Csp2 is not a CDK
> atom type name... the CDKHueckelAromaticityDetector expects both CDK
> atom type names and IAtomType.hybridization to be set properly...
I don't know what is wrong with the atom types, but I get this error when 
running the fingerprinter. Note there is a bug #1965254 which seems to point 
to the same problem.
> How many fingerprints do you now test? Can we rely on your new unit
> test to 'notice' when we need to recalculate the index?
10. No idea really if it is enough. How to find out?