Re: [Cdk-devel] FUCKING MODEL BUILDER

[Rajarshi G. <rg...@in...>]

On Sep 24, 2008, at 10:51 AM, Stefan Kuhn wrote:

> I always assumed it's the
> responsibility for those who do changes to make sure there are no side
> effects. I THINK THIS SHOULD ALSO BE TRUE FOR THE MODEL BUILDER.

I agree

> In case
> somebody feels like doing about the current problem, here's a hint:  
> What
> fails right now are the aromatic compounds. For them, the  
> fingerprints in the
> template library and of the molecule to lay out are different. The  
> reason for
> this seems to be the fingerprinter using atom types. In both cases  
> all atoms
> are set to C and all bonds to single to have a fallback. But since  
> atom
> typing is done before laying out, the atom types are still in. And  
> it looks a
> bit like this is causing differences in fingerprints.

Well my changes to the fingerprinter did not affect atom typing at  
all - the only change was how the paths are discovered.

The result is that the new fingerprints differ in the order of the  
atom/bond symbols (sometimes it is reversed from what the old  
fingerprinter would generate).

In the fingerprinter JUnit test, every test except one that  
explicitly looked at bit positions passed. Obviously the test that  
looked at explicit bit positions was updated for the new bit positions.

I don't know much about this module but I can see why  
TemplateHandler3DTest.testFingerprints will fail, since it is looking  
at explicit bit positions.

I can certainly appreciate your frustration, but I did send mail to  
the list asking how much of an impact the new fingerprinter would  
have on other code and it seemed that it was not a big issue.

-------------------------------------------------------------------
Rajarshi Guha  <rg...@in...>
GPG Fingerprint: D070 5427 CC5B 7938 929C  DD13 66A1 922C 51E7 9E84
-------------------------------------------------------------------
Chemistry professors never die, they just fail to react.