Clustering of Chemical Compounds using Unsupervised Neural Networks Algorithms: a comparison

Abstract

Clustering of chemical databases has tremendous significance in the process of compound selection, virtual screening and in the drug designing and discovery process as a whole. Traditionally, hierarchical methods like Ward’s and Group Average (Gave) and nonhierarchical methods like Jarvis Patrick’s and k-means are preferred methods to cluster a diverse set of compounds for a number of drug targets (using fingerprints based descriptors). In this work the applications of a number of self-organizing map (SOM) neural network algorithms to the clustering of chemical datasets are investigated. The results of the SOM neural networks, Wards and Group-Average methods are evaluated for the clustering of different biologically active chemical molecules that can be used as drug like compounds based on topological descriptors. The results show that the Wards and Group Average methods are equally good; however, the performance of Kohonen neural selforganizing maps (SOM) is also important due to its almost similar performance as the hierarchical clustering methods with the advantage of its efficiency.