Machine learning‐driven blood transcriptome‐based discovery of SARS‐CoV‐2 specific severity biomarkers

Abstract The Coronavirus disease 2019 (COVID‐19) pandemic, caused by rapidly evolving variants of severe acute respiratory syndrome coronavirus (SARS‐CoV‐2), continues to be a global health threat. SARS‐CoV‐2 infection symptoms often intersect with other nonsevere respiratory infections, making early diagnosis challenging. There is an urgent need for early diagnostic and prognostic biomarkers to predict severity and reduce mortality when a sudden outbreak occurs. This study implemented a novel approach of integrating bioinformatics and machine learning algorithms over publicly available clinical COVID‐19 transcriptome data sets. The robust 7‐gene biomarker identified through this analysis can not only discriminate SARS‐CoV‐2 associated acute respiratory illness (ARI) from other types of ARIs but also can discriminate severe COVID‐19 patients from nonsevere COVID‐19 patients. Validation of the 7‐gene biomarker in an independent blood transcriptome data set of longitudinal analysis of COVID‐19 patients across various stages of the disease showed that the dysregulation of the identified biomarkers during severe disease is restored during recovery, showing their prognostic potential. The blood biomarkers identified in this study can serve as potential diagnostic candidates and help reduce COVID‐19‐associated mortality.