Leveraging Explainable Automated Machine Learning (AutoML) and Metabolomics for Robust Diagnosis and...

Area of Science:

• Biomarker Discovery
• Computational Biology
• Metabolomics and Lipidomics

Background:

• Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a complex, debilitating illness with unknown causes and no objective diagnostic markers.
• Current diagnostic approaches for ME/CFS rely on symptom assessment, highlighting the need for objective biomarkers.
• Plasma metabolomic and lipidomic profiling offers a promising avenue for identifying ME/CFS-specific molecular signatures.

Purpose of the Study:

• To develop and validate an accurate diagnostic tool for ME/CFS using Automated Machine Learning (AutoML).
• To analyze plasma metabolomic and lipidomic data for identifying discriminatory biomarkers of ME/CFS.
• To gain biological insights into ME/CFS pathophysiology through interpretable AI models.

Main Methods:

• Utilized a public dataset of 888 metabolic and lipidomic features from 106 ME/CFS patients and 91 controls.
• Benchmarked three AutoML frameworks (TPOT, Auto-Sklearn, H2O AutoML) for ME/CFS classification.
• Employed univariate ROC, PLS-DA, cross-validation, permutation testing, and SHAP analysis for feature selection and model interpretability.

Main Results:

• The TPOT AutoML framework achieved high diagnostic performance: 92.1% AUC, 87.3% accuracy, 85.8% sensitivity, and 89.0% specificity.
• PLS-DA confirmed statistically significant discrimination between ME/CFS patients and controls.
• Explainable AI identified key metabolites linked to mitochondrial dysfunction, inflammation, gut-brain axis, and cell membrane integrity.

Conclusions:

• AutoML, particularly TPOT, provides a highly accurate and robust method for ME/CFS detection using omics data.
• The identified biomarkers offer valuable insights into ME/CFS pathophysiology, potentially guiding clinical decision support.
• This approach paves the way for developing novel therapeutic targets for ME/CFS.