Análise Multivariada de Dados em Petroleômica por Técnicas de Alta Resolução: Espectrometria de Massas de Ressonância Ciclotrônica de Íons por Transformada de Fourier e Ressonância Magnética Nuclear

• Autor(a) principal: Folli, Gabriely Silveira
• Data de Publicação: 2024
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Espírito Santo (riUfes)
• Texto Completo: http://repositorio.ufes.br/handle/10/17321
• topic: Área(s) do conhecimento do documento (Tabela CNPq); Alta resolução; dados complexos; sensibilidade; precisão; perfil espectral; correlação; quimiometria; petroleômica; aprendizado de máquina; subject.br-rjbn

Crude oil is a complex matrix, and the more in-depth study of its chemical structure (Petroleomics) has begun to be adopted by high-resolution techniques such as Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR MS) and high-field nuclear magnetic resonance (NMR) spectroscopy. However, high-resolution data presents challenges in spectral processing due to spectral variations. Consequently, the objective of this thesis was to develop multivariate data analysis applications (classification, regression, and design of experiments) to help overcome some of the limitations posed by high-resolution data. The first objective was to investigate the spectral profiles of analytical instruments (NMR, FT-ICR MS, nearinfrared – NIR, mid-infrared – MIR, and high-efficiency gas chromatography – HTGC). It was observed that high-resolution spectra predominantly exhibited a discrete profile (less correlated variables), while lower-resolution spectra showed a more continuous profile (more correlated variables), making them better suited for information clustering methodologies. The second objective involved estimating the intermediate precision of high-resolution mass spectrometers (FT-ICR MS and Orbitrap MS). It was noticed that both equipment presented oil with similar classes, but FT-ICR MS attributed a higher number of statistically assigned signals and had a lower detection limit compared to Orbitrap MS, due to its higher sensitivity. Furthermore, repeatability and intermediate precision of both spectrometers, although similar, demonstrated better values for FT-ICR MS in comparison to Orbitrap MS. The third objective focused on optimizing experimental parameters for ESI(±)FT-ICR MS in crude oil analysis. Plackett-Burman filtering planning was used to identify significant parameters, and the optimal analysis conditions were determined through a full factorial design. Global desirability determined by spectral quality metrics served as the response parameter for all experimental designs. The fourth objective aimed to develop a variable selection method that identifies variable correlations (Angular Search Algorithm with Variance Inflation Factor – ASA-VIF) and applies it to high-field 1H NMR, comparing it with MIR and NIR in linear (Partial Least Squares, PLS) and non-linear (Support Vector Regression, SVR) regression. An outlier identification methodology for non-linear models was also created. The results demonstrated that 1H NMR performed better using ASA-VIF-SVR. Furthermore, this selection drastically reduced the amount of information in MIR and NIR compared to 1H NMR, as these instruments contained a greater amount of correlated information (see the first application chapter). The final objective involved constructing a methodology for generating virtual samples (synthetic samples and artificial outliers) in complex data group classification models. This was essential as sample balance is crucial for more reliable metrics. The models showed good performance using virtual samples in both linear (PLS-DA) and non-linear (SVR) methodologies and with highresolution (Orbitrap MS) and low-resolution (MIR) spectra. In conclusion, all objectives resulted in improvements in the treatment of complex data using highresolution techniques.