Drowning in diversity? A systematic way of clustering and selecting a representative set of new psychoactive substances

Abstract

New psychoactive substances (NPS) can be generally described as a set of compounds that have been designed to mimic the effects of illegal recreational drugs, but are not subject to restriction or control with respect to existing regulations and legislation. In recent years, the number and chemical diversity of emergent NPS has increased substantially, and regulators have struggled to develop methods for accurate detection of NPS at the same rate. Existing approaches to NPS classification are pragmatic and/or semi-systematic and do not lend themselves to objective spectroscopic classification of emergent NPS. As such, this research discusses the identification of a systematic NPS classification based on chemical structures. A set of 478 NPS were grouped according to the similarity between their chemical structural features using hierarchical clustering and a maximum common substructure of 9 atoms, which included both hydrogen and heavy atoms. The rationale for including hydrogen atoms is that accurate spectroscopic identification of NPS will be dependent upon variations in substitution patterns in the molecules. This analysis generated 79 clusters, arising from 21 superclusters. The medoid substances of each cluster were used to form a dataset that was representative of the chemical space encompassed by known NPS. Subsequent categorisation of a test set of NPS showed that the test substances were assigned to an appropriate cluster when the Tanimoto similarity coefficient between the cluster medoid and the test substance was at least 0.5. This indicates that the cluster medoids could be used for assignment of emerging NPS to systematically-defined categories based on chemical structure. These medoids will also aid in the prediction of spectroscopic properties for emergent NPS, which will be invaluable for structure-based classifications and development of methods for detection of emerging NPS.