Chemometric Comparison Of GC-MS And GC-VUV For The Trace Analysis Of Methamphetamine

Abstract

Chemometrics, the application of mathematical or statistical algorithms to make inferences on the state of a chemical system from physical measurements of it, is a powerful tool that can be used to re-read what previously was observed as ‘noise’ in analytical measurements. Instruments such as spectrophotometers can take thousands of measurements over a predefined interval, but the spectra are only of great use when reference libraries exist, or if large trends occur that allow for visual matching, such as with a particular functional group. Application of statistical techniques to these data, such as principal component analysis (PCA) and linear discriminant analysis (LDA), can help to spot underlying variances, and differentiate between similar spectra by using linear combinations of these variables for classification. Methamphetamine (MA) is a member of the phenethylamines, a group of compounds that act as central nervous system stimulants, which are highly addictive and often the subject of law enforcement efforts at the local and federal level. Use of derivatization agents in analysis of seized narcotics is common practice, as it increases volatility/thermal stability of analytes, and improves peak shape for chromatographic resolution. In this analysis, we looked to investigate the difference in instrumental response for MA in its native form, as well as derivatized with two common agents, acetic anhydride and trifluoroacetic anhydride. These three forms were analyzed both on a gas chromatograph- mass spectrometer (GC-MS) and a gas chromatograph- vacuum ultraviolet spectrometer (GC-VUV). The raw GC-MS and GC-VUV data were separately normalized, and the dimensionality of the data was reduced through PCA, which uses orthogonal linear transformations of the data to capture most of the variance between datasets while simultaneously reducing the dimensionality for further analysis. Linear discriminant analysis was utilized to look at the principal components from PCA, and a classification model was built for use in discriminating between forms of methamphetamine from compressed datasets.