Abstract
Vibrational spectroscopy (infrared and Raman spectroscopy) is based on the interaction of matter with electromagnetic radiation. These techniques probe a wide range of information including physical and chemical properties yielding large dimensions of data. The advancement in the art of chemometrics for accurate prediction of sample properties presented a crucial turning point in the development of the spectroscopic methods, facilitating the interpretation of the large spectral data.
This thesis investigates the application of different chemometric modelling techniques to vibrational spectroscopic data from samples like meat, pharmaceuticals and immune cells. The nature of information probed determined the choice of chemometric techniques; however, some techniques gave better approximation (performance) for one sample than the others might. The analytical applications conducted in this thesis includes; discrimination of red meat species, quality assessment of red meat using conventional Raman and infrared spectroscopy, discrimination of cancer and immune cells using Raman microscopy and isothermal dehydration of pharmaceutical crystalline hydrates using low-frequency Raman spectroscopy.
Chapter one of this thesis introduces the concept of vibrational spectroscopy, spectroscopic instrumentation, spectral preprocessing, multivariate analysis (chemometrics) and highlights the aims of this thesis.
The second chapter investigates the discrimination of red meat species (beef, lamb and venison) using Raman and infrared spectroscopic techniques. Classification models were built using partial least square discriminant analysis (PLSDA) and support vector machine classification (SVMC); whereas principal component analysis (PCA) was employed for exploratory analysis. Visual assessment of the PCA scores revealed distinct separation of the three meat species for both spectroscopic techniques. Classification models built using the Raman data and validated against an independent test set yielded a classification accuracy ≥ 80 % and ≥ 92 % for the PLSDA and SVMC methods, respectively. Classification model created using the infrared data yielded an accuracy ≥ 94 % for both chemometric methods. This suggests that both Raman and infrared spectroscopic methods posit an effective tool for red meat discrimination.
Chapter three of this thesis investigates the implementation of Raman and infrared techniques as well as three data fusion strategies to evaluate pH and percentage intramuscular fat (% IMF) content of red meat. Quantitative models were built using partial least square regression and validated against an independent test set. Results obtained suggest a good correlation between the reference and predicted pH values using the Raman, infrared and high-level data fusion strategy whereas Raman and low-level fusion showed similar level of performance for predicting the % IMF content in red meat.
In chapter four, low-frequency Raman spectroscopy was shown to be very sensitive for monitoring the in situ isothermal dehydration of piroxicam and theophylline monohydrates. The dehydration was performed at four different temperatures and monitored in both the low-wavenumber (20 – 300 cm-1) and mid-wavenumber (335 – 1800 cm-1) Raman regions. Analysis performed using multivariate curve resolution (MCR) suggested the formation of specific anhydrous forms of piroxicam and theophylline upon dehydration of their respective monohydrates. The formation of the anhydrous forms was also detected on different timescales (approx. 2 min) between the low- and mid-wavenumber Raman regions. This finding highlights the differing nature of the vibrations being detected between these spectral regions.
In chapter five, Raman spectroscopy was demonstrated as a sufficient tool to discriminate cancer and immune cells. Phenotype of T-cells and monocytes were incubated with media conditioned by glioblastoma stem-cells (GSCs) showing different molecular background. Multivariate analysis performed using principal component - linear discriminant analysis (PCA-LDA) and SVM yielded sensitivities and specificities ≥ 70 % and ≥ 67 %, respectively. The results were in agreement with the flow cytometry analysis.