Screening for illicit psychoactive drugs based on pattern recognition methods

We are presenting an exploratory analysis assessing the feasibility of using pattern recognition methods for building an automated system screening in situ for illicit psychoactive drugs of abuse. The study is focused on ephedrine analogues, ergogenic aids which are also the main precursors of the most popular designer drugs, i.e. amphetamines. Each compound included in the training database was first characterized based on its infrared spectrum recorded between 1405 and 1150 cm−1. These spectra have been preprocessed with a feature weight, which enhances the absorptions that are the most specific to each of the modeled classes of compounds. The effect of two feature weights, wE2 and (wE-1)2, on the modeling and discrimination power of the system have been compared by using Principal Component Analysis (PCA) and Agglomerative Cluster Analysis (ACA). The dendrograms have been obtained based on the PCA scores of the modeled compounds. The influence of the number of principal components taken into account to model the targeted classes of illicit drugs is also discussed.