Métodos multivariados para a elucidação de informações analíticas em amostras complexas

Abstract

The study of complex samples such as food, drugs, and biological fluids require a robust data analysis methods, since these samples may contain information tens of thousands of constituents. The research developed in this work has as main theme the application of multivariate analysis methods to extract information from the vibrational spectra (infrared and Raman) or mass spectra of complex samples. Separation techniques such as CE, GC and LC are used as an auxiliary approach to providing information to the proposed multivariate models. Five applications are discussed in the text. The first makes use of PCA on GC chromatograms of samples of soybean oil and olive oil, searching for an exploratory analysis of the two classes of vegetable oils. Following are analyzed with mixtures of controlled additions of soybean oil in olive oil, by infrared absorption and Raman scattering techniques, with the aim of proposing a PLS regression to determine the percentage of addition. A second application addresses the search vibrational markers, infrared and Raman, that may be used to quantify the percentage of fat in milk fluid samples. In this study we carried out a profile analysis of fatty acids present in milk by CG technique, for the purpose of selecting a representative molecule of triacylglycerol was used to simulate vibrational spectra by theoretical models. From the spectra obtained by computer simulation was proposed assignment of vibrational modes present in milk fat. It was also proposed an MLR predictor model of the percentage of fat in fluid samples, whose vibrational variables were previously selected from a PLS model. Another application has been the use of PLS model of Raman spectra for the simultaneous determination of concentrations of rifampicin, isoniazid, ethambutol and pyrazinamide in tablet samples used in anti-tuberculosis treatment. The fourth application is focused on quantifying the percentage of whey added fraudulently in milk samples for analysis of the fatty acid profile by GC and CE. A discriminant analysis was used to select fatty acids markers that were monitored by MLR regression. Finally, it shows the preparation of a complete tutorial of univariate and multivariate data analysis, developed in R-software, for the determination of biomarkers in metabolomics experiments based on mass spectra analysis. All these applications have the intersection as the use of multivariate methods of analyzing data as the main tool to propose new vibrational / metabolite markers as well as alternative methods for quantification of various analytes.