Choosing relevant functional groups for optimizing Artificial Neural Networks detecting NBOMe hallucinogens

Abstract

In  the early 2010s,  a new group of  illicit psychedelic phenethylamines  was reported by the law enforcement agencies,  namely  the  NBOMe hallucinogens. The latter seem to be sold on the black  market as an alternative to LSD, due to their  powerful  psychoactive effects. The goal of  this study was to develop an  optimized Artificial Neural Network (ANN) able  to classify NBOMe hallucinogens based on their functional groups. These chosen molecular descriptors (functional groups) have been computed, by using the Dragon 5.5 program, for  the molecular structures of the main  NBOMe hallucinogens, which have been first optimized by using the Hyperchem program.  The  ANN system  was  built with the Easy NN plus program. Then, the importance of each functional group has been assessed. A new input database has been built with the functional groups found to be the most important.The performance of the new ANN system  has been characterized  based on  several classification accuracy criteria. The impact of the variable selection on the ANN performances  is discussed in detail.