Metabolomic Classification of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome via Explainable Ensemble Learning...

Area of Science:

• Biochemistry
• Computational Biology
• Immunology

Background:

• Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a complex multisystem illness lacking objective diagnostic biomarkers.
• Current diagnostic approaches are limited by the absence of established biomarkers, hindering effective clinical management.
• Plasma metabolomics offers a comprehensive view of biochemical alterations in ME/CFS but faces challenges in data interpretation.

Purpose of the Study:

• To evaluate ensemble classifiers, specifically Explainable Boosting Machine (EBM), XGBoost, and LightGBM, for binary classification of ME/CFS using plasma metabolomic and lipidomic data.
• To investigate the utility of feature dimensionality reduction techniques, such as Pareto-Guided Recursive Neural Network (PRNN), in enhancing classifier performance.
• To explore the interpretability of machine learning models in identifying key metabolic signatures and metabolite interactions indicative of ME/CFS.

Main Methods:

• Plasma metabolomic and lipidomic profiles from 197 participants (106 ME/CFS, 91 controls) were analyzed.
• Feature dimensionality was reduced using a Pareto-Guided Recursive Neural Network (PRNN) pipeline.
• Three ensemble classifiers (EBM, XGBoost, LightGBM) were trained and validated using 50-repeat stratified hold-out validation.

Main Results:

• Explainable Boosting Machine (EBM) achieved the highest classification accuracy (0.909) and AUC (0.940), outperforming XGBoost and LightGBM.
• Interpretability analysis identified significant contributions from pairwise metabolite interactions, including proline & indole-3-lactate and tyrosine & N-acetylornithine.
• Ablation analysis confirmed that metabolite co-variation, beyond individual levels, significantly enhances discriminative value, implicating amino acid metabolism and mitochondrial function.

Conclusions:

• EBM-based metabolomic profiling demonstrates robust internal validation for ME/CFS classification, offering a promising avenue for biomarker discovery.
• The study highlights the critical role of pairwise metabolite interactions in understanding ME/CFS pathophysiology and improving diagnostic accuracy.
• Explainable artificial intelligence (XAI) in metabolomics provides valuable insights into population-level metabolic signatures and individual patient heterogeneity.