Structure Based Clustering of NCI's Anti-HIV Library
To support clustering, new software called JKlustor has been developed as an add-on module for ChemAxon's chemical database handling system, JChem. The application can generate 2D hashed fingerprints for molecules, but real number descriptors may also be used during the calculation. The clustering process applies a version of the Jarvis-Patrick method (Jarp), which is based on variable-length neighbor lists. In the case of fingerprint input, the measure of similarity is the Tanimoto coefficient. Another clustering module, applying Ward's minimum variance method and using Murtagh's reciprocal nearest neighbor (RNN) algorithm as a heuristic, is also introduced.
EuroCombi-1, First Symposium of the European Society of Combinatorial Science, Budapest, July 1-5, 2001