Re: [Rdkit-discuss] Calculating MACCS Keys and default similarity metrics

[Shantheya B. <s.b...@tu...>]

Hey Greg,

thank you for your reply.


Am 17.11.2016 um 06:27 schrieb Greg Landrum:
> Hi Shanthy,
>
> On Wed, Nov 16, 2016 at 9:07 AM, Shantheya Balasupramaniam 
> <s.b...@tu... <mailto:s.b...@tu...>> wrote:
>
>
>     as far as I' ve seen there are two possibilites to calculate MACCSKeys
>     Fingerprints with RDKit.
>     Is there a principle difference between Maccskeys.GenMACCSKeys() and
>     AllChem.GetMACCSKeysFingerprint()?
>     Since I calculated the fingerprints for a couple of molecules with
>     both
>     options and the outcome was the same, I assume that it doesn't matter
>     which method is used?
>
>
> Correct, the function in rdkit.Chem.MACCSkeys just calls the same C++ 
> function that AllChem.GetMACCSKeysFingerprint() calls. It's still 
> there for backwards compatibility reasons.
>
>     Furthermore the default similarity metric for the Topological
>     Fingerprints and MACCS Keys is set to Tanimoto while for the Morgan
>     Fingerprints Dice is used.
>
>
> Where are you finding that default?

When I went through the RDKit Documentation I noticed that the default 
similarity metric used by DataStructs.FingerprintSimilarity() is 
Tanimoto and the examples for the Morgan Fingerprint incorporate Dice...

>     Is there a particular reason for the preference of the different
>     similarity metrics for the various Fingerprints?
>
>
> Not really. I used to by default choose Dice when I was working with 
> count-based fingerprints and Tanimoto when working with bit vectors. I 
> no longer worry too much about that. Some people probably still have 
> strong opinions on the matter; the RDKit supports whichever you 
> prefer. :-)

That's exactly the answer I hoped for :)
>
> -greg
Cheers,

Shanthy