Artificial neural networks applied for the identification of JWH cannabinoids based on structural descriptors

The goal of this study was to develop expert systems specialized in the automated recognition of JWH synthetic cannabinoids based on structural descriptors. The expert systems were built by using Artificial Neural Networks (ANN) and are designed to predict the class identity of the modeled compounds. The database consists of 160 compounds representing drugs of abuse (mainly synthetic cannabinoids, stimulants, hallucinogens, sympathomimetic amines, narcotics and other potent analgesics), precursors, or derivatized counterparts. Their molecular structure has been optimized and then characterized by 41 constitutional descriptors (CDs) and 34 functional group counts (FGs). Each of these two datasets has been used as input for two ANN systems, i.e. CD-ANN and FG-ANN, which distinguish between JWH compounds and negatives (non-JWHs). The validation results obtained for these two ANNs are presented and compared in detail, especially from the point of view of the classification accuracy. The potential use of these systems for predicting the toxicity of new JWH synthetic cannabinoids is also discussed.