K-means clustering analysis, ADME/pharmacokinetic prediction, MEP, and molecular docking studies of potential cytotoxic agents

In this paper, a combined approach based on a k-means algorithm and statistical analysis has been applied successfully to classify 500 cytotoxic agents using 21 molecular descriptors. The k-means algorithm applied on the first two principal components split the observations into homogeneous clusters, in order to examine similarities and dissimilarities between molecules within and between clusters. Three clusters are clearly distinguished and the percentage of molecules in each cluster is 50%, 24.88%, and 25.12% for cluster 1, cluster 2, and cluster 3, respectively. Silhouette analysis is used as a cluster validation method, confirming that all the molecules are very well clustered. The silhouette indices obtained for each cluster are 0.53 for cluster 1, 0.60 for cluster 2, and 0.57 for cluster 3. Therefore, the average silhouette index obtained is 0.56. The value of the Hopkins statistic obtained is 0.922 confirming that the dataset is highly clusterable. In addition, the paragons of each cluster have been identified. The test statistic was performed to characterize each cluster by a subset of molecular descriptors. Moreover, in silico screening of pharmacological properties ADME and evaluation of drug-likeness of molecules showed that cluster 1 molecules have the best ADME profile and drug-likeness. The quantitative analysis of molecular electrostatic potential (MEP) on van der Waals surface was performed to identify the nucleophilic and electrophilic sites in the paragons molecules. Finally, molecular docking was performed for the paragon molecules on six different targets. Docking studies support the results observed in the MEP analysis showing that the favorable reactive sites of molecules are involved in strong hydrogen interactions with the functional residues of receptors.