U.S. flag

An official website of the United States government, Department of Justice.

NCJRS Virtual Library

The Virtual Library houses over 235,000 criminal justice resources, including all known OJP works.
Click here to search the NCJRS Virtual Library

Comparison of Variable Selection Methods Prior to Linear Discriminant Analysis Classification of Synthetic Phenethylamines and Tryptamines

NCJ Number
254033
Journal
Forensic Chemistry Volume: 11 Dated: December 2018 Pages: 77-86
Author(s)
Amanda L. Setser; Ruth W. Smith
Date Published
December 2018
Length
10 pages
Annotation

This study developed and validated linear discriminant analysis (LDA) models for the characterization of compounds according to structural class based on mass spectral features.

Abstract

The Identification of emerging synthetic designer drugs is challenging in the absence of reference materials for subsequent mass spectral comparison. In the current study, a set of synthetic phenethylamines and tryptamines was analyzed by gas chromatography-mass spectrometry (GC-MS), and the resulting spectra were probed to identify characteristic variables (m/z values) for model development. Two approaches were used to select characteristic variables. First, ions known to be characteristic of each compound class were selected, resulting in a total of 13 variables that were used to develop an LDA model. Following cross validation of the model, a test set of phenethylamines and tryptamines was classified, resulting in a successful classification rate of 93 percent. In the second approach, principal components analysis (PCA) was used as a more objective method for variable selection. With this approach, a total of nine variables were selected and the resulting LDA model generated a successful classification rate of 86 percent. Although the classification success of each LDA model was similar, the PCA method for variable selection was substantially less time consuming than the more detailed approach that requires probing mass spectra for characteristic features. Classification models such as those reported by this study have potential utility for the class characterization of emerging analogs for which reference materials are not yet available. (publisher abstract modified)