Metabolomics-Based Machine Learning Diagnostics of Post-Acute Sequelae of SARS-CoV-2 Infection.
Cai, Ethan, Kouznetsova, Valentina L, Tsigelny, Igor F · Metabolites · 2025 · DOI
Quick Summary
Researchers developed a computer program that can identify Long COVID (PASC) by analyzing chemical markers in the blood called metabolites. The program successfully distinguished Long COVID from several similar conditions like Lyme disease and POTS, but had difficulty telling Long COVID apart from fibromyalgia, suggesting these two conditions may share similar chemical patterns in the body.
Why It Matters
This research provides potential support for a biological basis of Long COVID through metabolomic profiling and offers a novel diagnostic approach that could reduce diagnostic delays—a significant challenge in both PASC and ME/CFS. The noted metabolomic similarity between fibromyalgia and Long COVID is particularly relevant for ME/CFS patients, as these conditions share clinical features and diagnostic ambiguity.
Observed Findings
The machine-learning model identified PASC-dysregulated metabolites (p ≤ 0.05) distinguishable by molecular descriptors
The model achieved 0.8991 AUC-ROC accuracy on independent test sets
The model successfully differentiated PASC from ME/CFS, Lyme disease, POTS, and IBS in pairwise comparisons
The model failed to differentiate PASC from fibromyalgia, indicating high metabolomic similarity
Molecular descriptors enabled the model to function across diverse metabolite types without database-dependent limitations
Inferred Conclusions
PASC has a distinct metabolomic signature that can be computationally distinguished from several post-infectious and chronic symptom conditions
Fibromyalgia and PASC share sufficient metabolomic similarity to suggest either overlapping pathophysiology or common downstream biological consequences
Molecular descriptor-based machine learning offers a generalizable diagnostic approach that could expedite PASC identification by circumventing lengthy exclusion processes
Remaining Questions
Are the metabolomic similarities between PASC and fibromyalgia due to shared underlying mechanisms or independent convergent pathways?
Can this metabolomic signature be used for differential diagnosis in clinical settings, or does high fibromyalgia similarity limit practical diagnostic utility?
What This Study Does Not Prove
This study does not establish causation for any metabolite abnormalities—only association. It does not determine whether metabolomic patterns are primary disease mechanisms or secondary consequences of illness. The inability to distinguish fibromyalgia from PASC raises questions about whether the model identifies disease-specific biology or shared features among post-infectious/chronic conditions, and clinical validation in real-world diagnostic settings remains needed.