IDENTIFICATION AND CANCER SCREENING ANALYSIS USING MULTIVARIATE APPROACH AND BIOSPECTROSCOPIC TECHNIQUES
Cancer. Molecular fluorescence. ATR-FTIR. Mass spectrometry. Multivariate analysis.
This thesis reports the application of both infrared and molecular fluorescence spectroscopy, as well as mass spectrometry, combined with multivariate analysis techniques for classification of (pre-) cancerous lesions in different specimens. In a first study, excitation/emission matrices of molecular fluorescence were obtained for different cell lines (including 3T3, ARPE, HEK, HepG2, HeLa, HT-29 e 786-0) and classification models were built by using a combination of the algorithms OPLS and UPLS-DA. Correct classification indexes of 100% and 75% were obtained for both classes, Normal and Cancer, respectively. In addition, it was evaluated the influence of the antibodies anti-MMP-2 and anti-MMP-9 in the performance of the classification models. In the presence of the antibodies, the correct classification indexes were considerably improved reaching 100% for both classes, Normal and Cancer, using the algorithms OPLS/UPLS-DA. In a second study, the ATR-FTIR spectroscopy was applied to obtain the spectra of blood plasma of both healthy women (negative for intraepithelial lesion or malignancy, NILM) and women with cervical intraepithelial lesion (SIL) of low grade (LSIL) or high grade (HSIL), caused by HPV virus. Multivariate classification models were built, aiming a screening methodology for cervical cancer. The algorithms PCA-LDA/QDA, SPA-LDA/QDA and GA-LDA/QDA were applied as classification tools and their performance was evaluated. In general, the results obtained by GA-QDA were the most satisfactory, by using only chosen spectral variables that could be related to chemical groups of different biomolecules. The models GA-QDA correctly classified NILM vs. SIL with sensitivity and specificity around 67-94% e 82-94%, respectively. For NILM vs. LSIL, sensitivity and specificity values were about 67-94% e 82-94%, respectively. For NILM vs. HSIL, the sensitivity and specificity values were 76-97% e 73-100%, respectively. In the third study, mass spectrometry was applied to obtain the spectra of lipids extracted from blood plasma of women of NILM (n=42) and SIL (n=34) classes. Multivariate classification models were built by using the classifiers LDA, QDA and SVM. SVM-based models allowed to discriminate the classes with sensitivity and specificity values of 83.3# and 80.0% for NILM and SIL, respectively. Some possible lipids were associated to each class, such as prostaglandins, phospholipids, sphingolipids, Tetranor-PGFM and a hydroperoxide lipid. The results achieved in all studies highlight the potentiality of the spectroscopic and multivariate techniques as possible methodologies for cancer screening, what could effectively contribute to reduce morbidity and mortality caused by cancer.