Minimum Path length

User fdee5ee126

20-12-2007 15:46:11

Hi all,


just a simple question:


Is there a posibility to set a minimum pathlength for the CF?


I searched the docu as well as the forum, but did not find any post.


However it should be no problem to implement such an option.


Has anyone already done that?





Cheers,


Markus

ChemAxon efa1591b5a

20-12-2007 17:58:32

Hi,





possible yes, but in what circumstances? I mean:





- in general use the corresponding XML configuration file (e.g. one smaple cfp.xml is located in the examples/config folder under your jchem installation directory), this should work if you are running screen,


- in your application, that is, via the API: http://www.chemaxon.com/jchem/doc/api/chemaxon/descriptors/CFParameters.html#setLength(int),


- in command line (generatemd): http://www.chemaxon.com/jchem/doc/user/GenerateMD.html#options, or simply run
Code:
generatemd -x
.





Does this help?





regards,


Miklos

User fdee5ee126

03-01-2008 10:38:16

Hi Miklos, sorry for the late answer,





First of all, I whish you all a very happy new year!





The reason for my question is that for daylight FPs there seem to be significant differences in performance for 0-7 vs 3-10 patternlength.


I read that in a paper (PubMedID: 17591764) and just wanted to conduct a


test with the Chemaxon FPs to get an impression of any effect regarding my dataset.


Therefore a simple flag for generatemd would suit my aims.


Do you have any experience with that phenomenon?


Thanks in advance,


Cheers Markus

ChemAxon efa1591b5a

03-01-2008 10:54:16

Hi Markus,





according to our experience with tweaking these parameters pattern length ranging from 5 to 7 give the best results, the corresponding bit count is 1 or 2 for larger pattern lengths.





5 is usually suitable for substructure search pre-filtering, while rigorous similarity search performs better with 7, though that also needs longer fingerprint like 1024 bits or more to avoid too dark fingerprints.





Bit count over 2 usually does not improve quality.





We are very curious to learn your results on your dataset, we would highly appreciate if you could share those with us and the community.





Best regards and a happy new year!





Miklos