On 28 September 2010 14:45, Noel O'Boyle <baoilleach@...> wrote:
> On 28 September 2010 14:42, Egon Willighagen <egon.willighagen@...> wrote:
>> On Tue, Sep 28, 2010 at 3:36 PM, Noel O'Boyle <baoilleach@...> wrote:
>>> Ok - first problem.
>>> de = cdk.qsar.DescriptorEngine(cdk.qsar.DescriptorEngine.MOLECULAR)
>>> for desc in de.getDescriptorInstances():
>>> spec = desc.getSpecification()
>>> descclass = de.getDictionaryClass(spec)
>>> One of the new descriptors returns None for descclass - all of the
>>> others return things like ("topological descriptor",). It's
>> Is that the only descriptor not classified, or are there others?
> That's all.
>>> I can workaround this without any problems, but I guess that it's a
>>> documentation bug on the CDK side.
>> I would say this is a bug indeed...
Another difference is that the path-based fingerprint seems to have
been bug-fixed between CDK 1.2.5 and 1.3.6. I have two test molecules
CCCC and CCCN, and I calculate the bits set in their binary
fingerprints. In CDK 1.2.5, all 4 of the bits of CCCC are present in
CCCN. In CDK 1.3.6, only three of CCCC's 4 bits are present.
I've just done it out on paper, and the CDK 1.3.6 results make more
sense. The 4th bit of CCCC is the C-C-C-C path itself, which clearly
is not present in CCCN.