AbstractsBiology & Animal Science

Vibrational Spectroscopy and Multivariate Analyses for the Detection of Adulterated Weightloss Products

by Jeremy Rooney

Institution: University of Otago
Year: 0
Keywords: MIR; NIR; Raman; Principal Component Analysis; Support Vector Machines; Partial Least Squares Regression; Anorectics; Laxatives; Adulterants; Vibrational Spectroscopy
Record ID: 1302312
Full text PDF: http://hdl.handle.net/10523/5414


The feasibility of vibrational spectroscopic techniques for the detection and semi-quantification of anorectic and laxative pharmaceuticals used as adulterants in herbal medicines was investigated. The vibrational spectroscopic techniques used were mid-infrared (MIR), near infrared (NIR) and Raman. Each of these may allow routine screening of herbal medicines that would otherwise not undergo analyses at New Zealand's borders. All three techniques are rapid, cost-effective, non-destructive and can be automated. A high pressure liquid chromatography (HPLC) method was designed, validated and utilised to obtain reference values to allow multivariate models to be constructed from the spectroscopic measurements. FT-Raman spectroscopy was determined unsuitable during sample measurements due to strong emission and sample burning. Conversely, FT-MIR and FT-NIR were pursued further. Principal component analysis (PCA) models showed separation of scores based on adulteration for the MIR and NIR datasets. Consequently, support vector machine (SVM) models were formed from the training sets' PCA scores and the performance of these models was then gauged by test set prediction. The accuracies for SVM classification ranged between 97-100%. Lastly, MIR and NIR training sets were used for the creation of partial least squares (PLS) models for semi-quantification of the two adulterants (phenolphthalein and sibutramine hydrochloride monohydrate). The test set predictive performances for the MIR models were high and were apposite for semi-quantification of the adulterants necessary to determine consumer risk; the NIR models showed a lesser performance and had larger root mean squared error of prediction (RMSEP) values. It was concluded MIR spectroscopy, when combined with PCA, SVM and PLS, is suitable for detection and semi-quantification of the adulterants in herbal medicines.