PICRUST ANALYSIS AND MACHINE LEARNING APPROACHES TO COVID-19

The maintenance of life on Earth depends on interactions among humans, animals, and the environment, which are central to the One Health concept. Anthropogenic changes can alter these interactions and contribute to disease emergence, such as the Coronavirus disease 2019 (COVID-19). Several studies have indicated that the gut and nasopharyngeal microbiota compositions of patients with COVID-19 differs from that of healthy individuals and may be associated with the mechanisms and stages of severity in COVID-19. An extensive analysis of a large microbiome of different local and human body regions is essential for elucidating the effects of the microbiome and COVID patients . Therefore, this study aims to characterize microorganisms and metabolic pathways in publicly available metabarcoding datasets and to apply bioinformatics and machine learning approaches to identify microbial and genetic signatures associated with biological or clinical conditions. As a preliminary analysis, publicly available microbiome data from 209 patients with COVID-19 were evaluated to investigate the association between microbiota composition and disease severity. Clinical data, such as comorbidities, and microbiota data, including operational taxonomic units (OTUs), were integrated and analyzed using dimensionality reduction and supervised machine learning, aiming to predict the infection severity. Random Forest was used to classify patients into mild, moderate, and severe disease groups and to evaluate the importance of the variables used for classification. The data were divided into a training set and a test set at an 80/20 ratio. The model achieved 95% accuracy, correctly classifying 38 of 40 test samples, with a Kappa coefficient of 0.9226. The moderate and severe groups were correctly classified, while two samples from the mild group were classified as moderate. These results suggest that microbiota-derived features can be used to identify patterns associated with COVID-19 severity. In addition, we will predict the metabolic pathways using PICRUST2. Based on these preliminary results, the next stage will expand the analyses to additional metabarcoding datasets and other machine learning approaches.