Combining Multiple K-Means Clusterings of Chemical Structures Using Cluster-Based Similarity Partitioning Algorithm

Consensus clustering methods have been used in many areas to improve the quality of individual clusterings. In this paper, graph-based consensus clustering, Cluster-based Similarity Partitioning Algorithm (CSPA), was used to improve the quality of chemical structures clustering by enhancing the ability to separate active from inactive molecules in each cluster and improve the robustness and stability of individual clusterings. The clustering was evaluated using Quality Partition Index (QPI) measure and the results were compared with the Ward's clustering method. The chemical dataset MDL Drug Data Report (MDDR) database was used for experiments. The results obtained by combining multiple K-means clusterings showed that graph-based consensus clustering, CSPA, can improve the quality of individual chemical structure clusterings.