Metabolomic Profiling of COVID-19 Using Serum and Urine Samples in Intensive Care and Medical Ward Cohorts

Abstract

The COVID-19 pandemic remains a significant global health threat, with uncertainties persisting regarding the factors determining whether individuals experience mild symptoms, severe conditions, or succumb to the disease. This study presents an NMR metabolomics-based approach, analyzing 80 serum and urine samples from COVID-19 patients (34 intensive care patients and 46 hospitalized patients) and 32 from healthy controls. Our research identifies discriminant metabolites and clinical variables relevant to COVID-19 diagnosis and severity. We propose a three-metabolite diagnostic panel—comprising isoleucine, TMAO, and glucose—that effectively discriminates COVID-19 patients from healthy individuals, achieving high efficiency. Recognizing that serum profiles are more reliable but invasive compared to urine samples, we propose reconstructing serum profiles using urine ¹H NMR data. Our robust multi-output regression model demonstrates high accuracy in this reconstruction, and in classifying the converted serum spectroscopic profile. This suggests the feasibility of determining COVID-19 infection and predicting its severity using a non-invasive sample such as urine.