Re: [Cdk-devel] Cdk-devel Digest, Vol 3, Issue 6

[<Mar...@ep...>]

The EStateAtomTypeMatcher code on CDK is currently incorrect which is
why the ALOGP doesnt work.  I have tested it on a dataset of nearly
13000 chemicals (**with different code used to determine the EState
fragments**).  The CDK prediction gives an R2 correlation coefficient of
0.84 with experimental logKow values and an R2 value of 0.986 for CDK
ALogP vs Dragon ALogP.  Dragon ALOGP gives an R2 value of 0.83 with the
experimental values so the slight difference with dragon doesnt
negatively impact the quality of the predictions.


> hi,
> I validated the AlogP descriptor commited by Martin which is in the
> svn (updated today in the morning). I used the ZINK testset with
> Hydrogen and without Hydrogen and an AlogP implementation which was
> supervised by Arup Ghose himself (A==Arup).
>
>
> Sorry for the format, but the result is that there is no correlation.
>
> best regards
> Christian