Novel Strategy for Human Deep Vein Thrombosis Diagnosis Based on Metabolomics and Stacking Machine Learning

Area of Science:

• Biochemistry
• Computational Biology
• Medical Diagnostics

Background:

• Deep vein thrombosis (DVT) poses significant clinical challenges, impacting patient morbidity and mortality.
• Current diagnostic methods for DVT face limitations in accuracy and efficiency.
• Metabolomics and machine learning show promise for improving disease diagnosis.

Purpose of the Study:

• To investigate the synergistic potential of dual-platform metabolomics (GC-MS and LC-MS) for enhanced DVT diagnosis.
• To develop a precise and accurate diagnostic model for deep vein thrombosis using advanced computational techniques.
• To create a user-friendly application for automated DVT diagnosis in clinical settings.

Main Methods:

• Serum samples from DVT patients were analyzed using both gas chromatography-mass spectrometry (GC-MS) and liquid chromatography-mass spectrometry (LC-MS).
• Differential metabolites were identified, and SHapley Additive exPlanations (SHAP) were used for feature engineering.
• A stacking machine learning model was developed and validated for DVT diagnosis.

Main Results:

• Sixty-one differential metabolites were identified, with 22 detected by GC-MS and 39 by LC-MS.
• Five key metabolites were selected using SHAP for model development.
• The developed stacking model demonstrated high diagnostic accuracy for deep vein thrombosis.

Conclusions:

• Dual-platform metabolomics significantly expands metabolite detection for DVT analysis.
• The integration of metabolomics and machine learning, particularly a stacking model, offers a powerful approach for accurate DVT diagnosis.
• A practical, user-friendly application system was created to facilitate clinical adoption of the DVT diagnostic model.