Li et al. (2026)
- Authors: Junrong Li, Hanyu Cao, Zirun Zhu, Kefeng Li
- Institutes: Fudan University, Shanghai, China, Medical Engineering Fund of Fudan University, Shanghai, China
- Publisher: Computational Biology and Chemistry
- Link: DOI
Summary
This study marks a significant advance toward a routine, objective clinical test for ME/CFS by identifying 10 key blood biomarkers from the massive UK Biobank dataset. The researchers developed a highly accurate diagnostic model that can distinguish ME/CFS from both healthy individuals and those with overlapping conditions. The discovery of causal links between metabolic markers like glucose and leucine and symptom severity provides new mechanistic insights that could guide future personalized treatments.
What was researched?
The study aimed to develop a lightweight and interpretable machine learning model for the objective diagnosis of ME/CFS using blood biochemical and metabolomic data.
Why was it researched?
Diagnosis currently relies on subjective clinical criteria, which leads to high rates of underdiagnosis and difficulty in differentiating ME/CFS from other conditions with similar symptoms.
How was it researched?
Researchers analyzed data from 1,137 ME/CFS cases and 66,838 controls in the UK Biobank using a pipeline that compared 11 machine learning models and various imputation and feature selection methods. The final model utilized 10 primary biomarkers alongside covariates such as age, BMI, and gender, and was interpreted using SHAP analysis and Mendelian randomization.
What has been found?
The optimized model achieved 93.9% accuracy and a ROC-AUC of 0.979 in identifying ME/CFS patients. Ten key biomarkers were identified, with urea, total protein, glucose, total bilirubin, leucine, and vitamin D showing established causal relationships. Notably, elevated levels of glucose and leucine were found to exacerbate CFS symptoms.
Discussion
A major strength of the study is the inclusion of controls with overlapping conditions, which increases clinical relevance. The use of explainable AI techniques like SHAP provides a transparent way for clinicians to understand how the model reaches a diagnosis.
Conclusion & Future Work
A lightweight diagnostic tool using routine objective biomarkers is both feasible and highly effective for ME/CFS. This model provides a practical framework for integrating objective diagnostics into clinical settings to improve patient outcomes.