Artificial Neural Networks designed to identify NBOMe hallucinogens based on the most sensitive molecular descriptors

Abstract

Drugs are largely responsible for a high percentage of the public health problems and drug-related mortality worldwide. The illicit drug market is constantly evolving in order to circumvent controls and confiscations. NBOMe is a newly discovered class of hallucinogenic drugs. Their representative synthetic derivative is 25I-NBOMe, which is sold on illicit channels in the form of powder or blotter paper. This study presents and compares a series of Artificial Neural Networks (ANN) designed to screen for NBOMe hallucinogens. The database is composed of 160 illicit drugs, such as NBOMe hallucinogens, narcotics, sympathomimetic amines and other potent analgesics, as well as some of their major precursors. The molecular structures of the main NBOMe hallucinogens were first optimized using the Hyperchem 8.03 software. Then, topological, 3D- MoRSE and constitutional descriptors, as well as functional groups of these compounds have been determined by using the Dragon 5.5 software. Then ANNs have been built by using only the most selective descriptors. Their accuracy was compared based on several figures of merit. The impact of variable selection on ANN performance and is also analyzed in detail.