"Identification of newly emerging synthetic designer drugs is challenging for forensic drug analysts due to the lack of available reference materials for a visual comparison of mass spectra. The focus in this work was the classification of synthetic drugs according to class or subclass using multivariate statistical procedures, specifically linear discriminant analysis (LDA). Reference materials representative of tryptamines and phenethylamines were analyzed by gas chromatography-mass spectrometry with a single quadrupole mass analyzer. Before LDA models could be developed, variable reduction was necessary. Two methods of variable selection were used. The first method used principal components analysis (PCA) as an objective approach and the second method used an informed chemical approach where mass spectra were probed for ions characteristic of each class or subclass of compounds. Ultimately, two variable sets were compared for classification success rates: the variable set selected by PCA (including nine m/z values) and the variable set selected using the informed chemical approach (including 13 m/z values). Two models were defined using the different variable sets and a common training set. A test set was then introduced to the model for classification. The LDA model using the informed chemical approach performed better, with a 93% classification success rate as opposed to the 86% success rate observed when using the variables selected by PCA. Overall, this research provides a classification procedure for compounds not identifiable by standard methods of comparison to a known reference material."--Page ii.
Read
