Random Forest Approach to Identifying Microbial Biomarkers Relating to COVID-19 Severity

The Coronavirus Disease 2019 (COVID-19) pandemic, caused by the novel coronavirus SARS-CoV-2, has led to significant mortality and hospitalizations worldwide. Although the spread of the pandemic has recently slowed, the persistence of long-COVID remains a continuous public health concern. Sub- stantial data have been amassed to delineate and synopsize the fundamental symptoms of COVID-19, encompassing respiratory distress, muscular discomfort, anosmia, limb paresthesia, and fatigue. Notwithstanding the extensive investigations conducted in this domain, a pivotal unresolved issue pertaining to the pandemic is the considerable interindividual variability in the severity of COVID-19 symptoms following SARS-CoV-2 infection. Despite concerted research efforts, the precise biomarkers from the human microbiome that underlie the cause of different severity of COVID-19 symptoms remain elusive. This study aims to develop a Machine Learning (ML) model for COVID-19 severity classification between asymptomatic and moderate and to identify critical microbial biomarkers for COVID-19 severity using previously published 16S rRNA gene sequencing data. The employed Random Forest (RF) model has exhibited outstanding performance with an accuracy of 0.82, an F1-score of 0.88 and an AUC of 0.83. Our results show Selenomonas, Neisseria, Prevotella, Veillonella and Rothia are critical microbial biomarkers related to COVID-19 severity.