LibMCS algorithm

User ec8f22af42

12-07-2011 07:59:42

Are there any further information about the algorithm used in the current version of LibMCS apart from this poster (

What kind of fingerprints are used to generate the similarity matrix?

Which similarity threshold is chosen? Which similarity metric (tanimoto?)?



ChemAxon efa1591b5a

14-07-2011 14:59:15


I'm afraid there's no other publication or white-paper available. 

We used ChemAxon's hashed binary fingerprint, perhaps ECFP that we've implemented recently would be more relevant and could do a better job.

Tanimoto is fine, though we found that the common bit count is more related to the MCS.

And finally, there's no direct similarity threshold used, as that can vary with respect to the particular structure set being cluster. Instead, we store a certain number of elements of the similarity matrix, regardless of their similarity scores.

Hope this helps!

Please tell us more about your use case and expected results!