Chemical profiling and classification of illicit heroin by principal component analysis, calculation of inter sample correlation and artificial neural networks

Artificial neural networks (ANNs) were utilised to validate illicit drug classification in the profiling method used at "Institut de Police Scientifique" of the University of Lausanne (IPS). This method established links between samples using a combination of principal component analysis (PCA) and calculation of a correlation value between samples. Heroin seizures sent to the IPS laboratory were analysed using gas chromatography (GC) to separate the major alkaloids present in illicit heroin. Statistical analysis was then performed on 3371 samples. Initially, PCA was performed as a preliminary screen to identify samples of a similar chemical profile. A correlation value was then calculated for each sample previously identified with PCA. This correlation value was used to determine links between drug samples. These links were then recorded in an Ibase (R) database. From this database the notion of "chemical class" arises, where samples with similar chemical profiles are grouped together. Currently, about 20 "chemical classes" have been identified. The normalised peak areas of six target compounds were then used to train an ANN to classify each sample into its appropriate class. Four hundred and sixty-eight samples were used as a training data set. Sixty samples were treated as blinds and 370 as non-linked samples. The results show that in 96 % of cases the neural network attributed the seizure to the right "chemical class". The application of a neural network was found to be a useful tool to validate the classification of new drug seizures in existing chemical classes. This tool should be increasingly used in such situations involving profile comparisons and classifications. (c) 2005 Elsevier B.V. All rights reserved.