This dissertation focuses on food characterization by chemometric methods applied to the high-performance liquid chromatography diode array detection (HPLC-DAD) measurements of Prunella L. species and honeys and attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIRS) measurements of gelatins.
Chapter 1 provides a review of chemometric methods which are especially useful for food analysis. Chemometrics, one of the subdisciplines of analytical chemistry, is widely used for complex measurements that are obtained from a variety of samples and applicable to a broad range of analytical methods. A list of chemometric and preprocessing methods are given in this chapter along with their descriptions. Super partial least square (sPLS), fuzzy rule building expert system (FuRES), support vector machines classification trees (SVMTreeG and SVMTreeH), sPLS discriminant analysis (sPLS-DA), linear discriminant analysis (LDA), bootstrapped Latin partitions (BLPs), and principal component analysis (PCA) were used in this work. Retention time alignment, normalization, the square root transformation, dissimilarity kernel, standard normal variate (SNV), and principal component orthogonal signal correction (PC-OSC) were applied as preprocessing methods.
Chapter 2 evaluates four different data representations of Prunella L. species using HPLC-DAD were evaluated to determine the total antioxidant activities. Prediction of total antioxidant activity of Prunella L. species was achieved by sPLS regression which was applied to the entire two-way chromatographic-spectral images, the average UV spectra, the total absorbance chromatogram, and the lambda max (¿max) chromatogram. Three different antioxidant assays, 2,2’-azino- bis(3-ethylbenzothiazoline-6-sulfonic acid) (ABTS), 2,2-diphenyl-1- picrylhydrazyl (DPPH) and Folin-Ciocalteu (FC), were used for measuring the total antioxidant activity with 12 different solvent systems. The coefficients of determination (R2) for the entire two-way chromatographic-spectral images (ABST (0.943±0.008), DPPH (0.91±0.01), and FC (0.963±0.006)) indicated good accuracy for predicting antioxidant activities. The entire two-way chromatographic- spectral images have been used for the first time for calibration.
Acidic hexane, as an extraction solvent, gave the least root mean square error of prediction (RMSEP) for the two-way chromatographic-spectral images, so it would be the best solvent for modeling antioxidant activities.
In Chapter 3, using the two-way images of phenolic compounds from HPLC-DAD, floral and chestnut honey from Turkey were successfully differentiated. FuRES, SVMTreeG, and sPLS-DA were used to develop classification models. Normalization, retention time alignment, square root transform, and dissimilarity kernel were evaluated as data preprocessing methods. The bootstrapped Latin partition (BLP) was used with 100 bootstraps and 4 partitions. Classification rates of FuRES and SVMTreeG with a square root transform were 97.6 ± 0.4% and 97.6 ± 0.4% for classifying the type of honey, respectively. These classifier also provided quite well classification rates of profiling honeys based on their geographical locations. The measures of precision are 95% confidence intervals. HPLC-DAD was demonstrated as a reliable analytical method for authentication of honey.
Chapter 4 describes the differentiation of bovine, porcine, and fish gelatins by their ATR-FTIRS spectra coupled with pattern recognition. Three tree-based classification methods, FuRES, SVMTreeG, and SVMTreeH, and one reference model, sPLS-DA, were evaluated with and without two preprocessing techniques, namely SNV and PC-OSC. Validation of these methods was obtained with 95% confidence intervals with 10 bootstraps and 4 Latin partitions (10:4). The ATR-FTIR spectra were used with four different ranges: full spectra (4000–650 cm–1), fingerprint region (1731–650 cm–1), specified spectra (4000–800 cm–1), and narrow fingerprint region (1731–800 cm–1). Classification rates for the methods were improved with SNV and PC-OSC when they were used separately or together. The highest classification rates were obtained from the narrow fingerprint region with SNV and PC-OSC at 97.4 ± 1.6% for FuRES, 100 ± 0% for sPLS-DA, and 99.3 ± 0.5% for both SVMTreeG and SVMTreeH. This research on gelatin profiling is the first to report classification rates to identify the animal source of gelatin using ATR- FTIRS. ATR-FTIRS combined with pattern recognition is a potential analytical technique for differentiating the sources of bovine, porcine, and fish gelatins with fast and reliable results.
Chapter 5 summarizes my work at OHIO and describes future work.