Predictive QSAR model and clustering analysis of some Benzothiazole derivatives as cytotoxic inhibitors

We propose an original approach dedicated to QSAR modeling and clustering analysis based on a dataset of 23 Benzothiazole derivatives as cytotoxic inhibitors. The choice of relevant molecular descriptors is a key step in QSAR modeling. In this work, model selection by exhaustive search is used to identify the best subset of molecular descriptors. Three distinct clusters have been identified using Kmeans clustering. Each cluster, groups a homogeneous class of molecules with respect their molecular descriptors. Silhouette analysis, used as cluster validation approach, proves that the molecules are very well clustered and there are no molecules placed in the wrong cluster. Moreover, the results emphasize that the molecular descriptors belonging to physico-chemical class appears to largely influence the cytotoxic activity of Benzothiazole derivatives. From this classification, all molecules with the trifluoromethyl group show a strong activity. The best cytotoxic activity was exhibited by compound containing two trifluoromethyl groups in ortho position. We suggested that this functional group is correlated to binding affinity. The PLS equations exhibit a good agreement between fitted and observed cytotoxic activities. According to the goodness of fit statistics, 67% and 80% of the variability in cytotoxic activity around its mean are explained. The Fstatistic test revealed the significance of both PLS regression equations. The values of bootstrapping correlation coefficient R 2