Hi Gonzalo, On Mon, Sep 7, 2015 at 11:19 AM, Gonzalo Colmenarejo-Sanchez < [email protected]> wrote:
> > > Is it posible to calculate Tversky similarities for pattern fingerprints? > It's definitely possible to do a Tversky similarity with the pattern fingerprints. They are the same as all the other RDKit fingerprints, so any of the built-in similarity measures can be used. > That would provide a more detailed idea of extent of substructure rather > than the categorical “yes” or “no” with AllProbesBitMatch. > In the case it were possible, do you know if that would result in more > accurate measures of extent of substructure than the regular fingerprints? > I'm not sure where you're trying to go with this. When pattern fingerprints are used for substructure screenout, it's important to use AllProbeBitsMatch() because that's a necessary (though not sufficient) condition for a substructure match to be present. If you are using them as a similarity measure, then I could see how using some type of asymmetric similarity measure - where Sim(A,B) != Sim(B,A) - could certainly be useful. -greg
------------------------------------------------------------------------------
_______________________________________________ Rdkit-discuss mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
