JKlustor, screen and descriptors API

This topic contains information about the new JKlustor, screen and descriptors API.

The attached code example uses the new descriptors / jklustor API to invoke a sphere exclusion clustering on input structures using ECFP fingerprints.

Notes on parts used from the APIs:

• Structure standardization is not part of the new descriptors API, in the example it is done using core API functionality


• Descriptor generation and comparison are represented by immutable classes (com.chemaxon.descriptors.fingerprints.ecfp.EcfpGenerator and com.chemaxon.descriptors.fingerprints.ecfp.EcfpComparator)


• The similarity space considered during the similarity based clustering algorithm (com.chemaxon.clustering.sphex.SphereExclusion in the current example) is decoupled from the input by com.chemaxon.clustering.common.DissimilarityInput interface


• The clustering result (represented by com.chemaxon.clustering.common.IDBasedSingleLevelClustering) therefore are not coupled to the actual input structure. The result represents integer IDs (indexes) of the associated DissimilarityInput


• A convenience class which encapsulate a set of molecule objects with an associated descriptor generator/comparators is provided, see com.chemaxon.clustering.common.MolInput and  com.chemaxon.clustering.common.MolInputBuilder

The attached code example uses the new descriptors / jklustor API to invoke a ward's minimum variance hierarchic clustering on input structures using ECFP fingerprints.

Additional notes besides the non-hierarchic (sphere exclusion) clustering example:

• The clustering result (represented by com.chemaxon.clustering.common.IDBasedHierarchicClustering) represents hierarchic clustering of integer IDs (indexes) of the associated DissimilarityInput.


• Levels are associated to the clusters through com.chemaxon.clustering.common.LevelAssigner


• LevelAssigner can provide partitioning of the associated clustering at a specified level. The partitioning is represented by a non-hierarchic clustering.