A New Clustering Algorithm Based on Pattern Extraction in Molecular Fingerprints

Abstract

In this paper an algorithm for the extraction of patterns in chemical fingerprints is described. As input this algorithm uses a fingerprint representation of the molecule dataset, generating a group of consistent disjoint patterns also represented as binary arrays, which are satisfied by not necessarily disjoint subsets of molecules in the dataset. The algorithm has been completely developed in Java, allowing its integration into free applications of computational chemistry. The algorithm has been tested, and the use of the patterns instead of the original fingerprints has presented an increase in the efficiency in the processes of datasets classification. The results show that it is possible to reconstruct the original fingerprints using the final group of patterns that characterize all the elements of the dataset.