Technological platforms in the prognosis of COVID-19: ATR-FTIR spectroscopy, metabolomic salivary analysis and machine learning

• Ano de defesa: 2022
• Autor(a) principal: Scheucher, Débora Lara Zuza
• Tipo de documento: Tese
• Tipo de acesso: Acesso embargado
• Idioma: eng
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Ciências da Saúde
• Palavras-chave em Português: COVID-19; ATR-FTIR; Metabolômica; Metabolomic; Aprendizado de máquina; Machine learning; Máquina de vetores de suporte; Support vector machine; Saliva; CNPQ::CIENCIAS DA SAUDE::MEDICINA::CLINICA MEDICA::DOENCAS INFECCIOSAS E PARASITARIAS; CNPQ::CIENCIAS DA SAUDE::FISIOTERAPIA E TERAPIA OCUPACIONAL; Ciências médicas; COVID-19 (Doença) - Prognóstico; Aprendizado do computador; Saliva - Prognóstico
• Link de acesso: https://repositorio.ufu.br/handle/123456789/37523; http://doi.org/10.14393/ufu.te.2023.7004

The COVID-19, caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, resulted in more than 636 million confirmed cases and 6.6 million deaths worldwide, and its management remains a challenge for health teams. The development of novel tools for predicting COVID-19 severity and prognosis could assist healthcare professionals in the decision-making process, and improve knowledge about the pathophysiology of the disease. In this context, attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectroscopy consists of a vibratory spectroscopy that can provide information about the structure and chemical composition of biological materials at the molecular level. Metabolomic analysis identifies unknown compounds, and determine structure and chemical properties of molecules, reflecting the metabolic state of organs and tissues. Moreover, machine learning is another promising platform that uses artificial intelligence to make the most appropriate decision based on pattern recognition within the analyzed data. Most of studies use these tools linked to the use of blood, however non-invasive methods are needed, for example, using saliva. This study aims to: develop predictive and prognostic value models of machine learning algorithms based on salivary infrared spectral signatures to the prognosis of (i) moderate vs severe COVID, (ii) discharge vs death, and (iii) no need of invasive mechanical ventilation (IMV) vs IMV outcomes for COVID-19 patients (Article 1); analyze the saliva metabolites from COVID-19 patients for disease stratification by (i) the disease progression - moderate or severe COVID, (ii) discharge vs death, and (iii) the need of IMV, performing the metabolomics of saliva collected in early stages of hospitalization (Article 2). We analyzed the saliva of 118 patients infected with SARS-CoV-2 by ATR-FTIR, machine learning and high-performance liquid chromatography hyphenated to the mass spectrometer (HPLC-MS) early in hospitalization (mean 3.0 ± 3.2 days). The classification of salivary infrared spectra by support vector machine (SVM) showed (i) a predictive value of severity – moderate vs severe [sensitivity of 68%, specificity of 52%, and accuracy of 60%], (ii) a predictive value of survival – discharge vs death [sensitivity of 60%, specificity of 64%, and accuracy of 64%], (iii) a predictive value of need of IMV (no IMV x IMV) [sensitivity of 46%, specificity of 72%, and accuracy of 66%] (Article 1). Metabolomic profile of COVID-19 patients showed an involvement of purines, amide, lipid, sulfoxide and ketones. Saliva metabolome allowed us to identify metabolites related to greater disease severity in the evaluated outcomes which might be used in selection of potential saliva biomarker for severity evaluation early in hospitalization and open new therapeutic perspectives against COVID-19 (Article 2). This study reveals the potential of salivary ATR-FTIR spectroscopy, metabolomic analysis and machine learning as prognostic technological platforms to determine outcome in COVID-19 patients.