Combining multiple classifications of chemical structures using consensus clustering

Abstract

Consensus clustering involves combining multiple clusterings of the same set of objects to achieve a single clustering that will, hopefully, provide a better picture of the groupings that are present in a dataset. This paper reports the use of consensus clustering methods on sets of chemical compounds represented by 2D fingerprints. Experiments with DUD, IDAlert, MDDR and MUV data suggests that consensus methods are unlikely to result in significant improvements in clustering effectiveness as compared to the use of a single clustering method.

Information Theory and Voting Based Consensus Clustering for Combining Multiple Clusterings of Chemical Structures

“…[24] If we assume that a cluster contains n compounds, that a of these are active and that there is a total of A compounds with the chosen activity. Then the precision, P, and the recall, R, for that cluster are: [16] P ¼ a=n ð3Þ…”

“…[37] It contains 17 activity classes, with each class containing 30 actives (total of 510 molecules). This dataset was used by Che et al [16] for consensus clustering experiments and also by Ammar et al for virtual screening experiments. [33][34][35] The MUV dataset details are shown in Table 2.…”

“…[14][15] For clustering of chemical structures, it is most unlikely that any single method will yield the best classification under all circumstances, even if attention is restricted to a single type of application. [16] Chu et al used some consensus clustering methods on sets of chemical structures and concluded that the consensus clustering methods can outperform Ward's method. [16] However, based on the implemented methods, it was not the case if the clustering is restricted to a single consensus method.…”

“…[16] Chu et al used some consensus clustering methods on sets of chemical structures and concluded that the consensus clustering methods can outperform Ward's method. [16] However, based on the implemented methods, it was not the case if the clustering is restricted to a single consensus method. In addition, Saeed et al [17] examined the use of graph-based consensus clustering methods, cluster-based similarity partitioning algorithm (CSPA) and hypergraph partitioning algorithm (HGPA), [18] for clustering of MDDR dataset and concluded that they can improve the effectiveness of individual clusterings and provide robust and stable clustering.…”

Chemoinformatics at the University of Sheffield 2002–2014

“…[1][2][3][4] The work of the chemoinformatics research group in Sheffield has always had a strong algorithmic and methodological focus, this reflecting our location in an informatics, rather than a chemical, academic department. We have thus drawn extensively on computational techniques from, e.g., graph theory, [5][6] cluster analysis, [7][8] image processing [9][10] and combinatorial optimisation [11][12] inter alia to design and implement a wide range of chemoinformatics applications. Lynch and Willett [1] and Bishop et al [4] have described Sheffield work in chemoinformatics for the periods 1965-1985 and 1986-2002, respectively.…”