Gil Sorribes, Manel2023-10-042023-10-042023-10-04http://hdl.handle.net/10230/58041Treball fi de màster de: Master in Intelligent Interactive Systems. Tutors: Mario Ceresa, Gabriele Leoni, Antonio Puertas, Mauro PetrilloPersonalized medicine is a rapidly evolving field to which many resources have been devoted recently. It represents a paradigm shift from a one-size-fits-all approach to healthcare, focusing instead on tailoring treatments and diagnoses to individual patients. This study aims to contribute to this transition by leveraging recent advancements in microbiome research. An earlier study that computed the Gut Microbiome Health Index (GMHI), a potent indicator capable of predicting disease presence with approximately 73% of accuracy, serves as the foundation for this research. The GMHI is a numerical value that classifies a person as healthy if the index is greater than zero, non-healthy if less than zero, and undetermined if the GMHI is equal to zero. The intent of this study is to further utilize and develop this index through two primary objectives. The first objective is to delve deeper into the dataset provided by the original GMHI study authors, using statistical and Machine Learning (ML) techniques. This process involves a variable analysis, that are species, and the application of classifier algorithms such as the Support Vector Machine. Upon determining the most informative variables, these will be used as input to a Neural Network (NN) in an attempt to surpass the existing GMHI prediction accuracy. Through leveraging the sophisticated pattern recognition capabilities of a NN, the aim is to refine the classification accuracy of health statuses, thereby providing more personalized diagnostic insights. Two different architectures were used: a Fully Connected NN and an Autoencoder NN. The results obtained from this first objective, though not exceptional, are promising. The accuracy slightly improved by approximately 3 percent when using both approaches, indicating better prediction of health conditions. The second objective is to apply the GMHI to a distinct dataset, which comprises information on individuals both affected and unaffected by COVID-19. This novel application of the GMHI represents an innovative investigation into the potential impact of a viral disease like COVID-19 on an individual’s gut microbiome. Given the worldwide repercussions of COVID-19, such exploration is crucial. As the potential implications of COVID-19 on the gut microbiome are largely unknown, this research could provide key insights, influencing patient treatment strategies and prognosis. Encouraging findings demonstrate that by rescaling and adapting the GMHI to this specific dataset, the accuracy in detecting a COVID-19 patient significantly improves by 17 percent, thereby achieving an accuracy of 75%, compared to the accuracy attained when directly using the authors’ formula. In summary, this study aims to integrate advanced ML techniques to enhance the GMHI and apply it to a novel dataset. This could significantly contribute to the burgeoning field of personalized medicine, where unique microbiome profiles may inform diagnostic and therapeutic strategies. The integration of computational techniques like ML could revolutionize the understanding of disease pathogenesis and treatment approaches, propelling healthcare into a new era. This project findings may aid in this direction and could serve as a baseline for further research, as outlined in the further steps section.application/pdfengTots els drets reservatsinfo:eu-repo/semantics/masterThesisGMHIGut MicrobiomeMetagenomicsMachine LearningNeural NetworkCOVID-19info:eu-repo/semantics/openAccess