Using an Interpretable Amino Acid-Based Machine Learning Method to Enhance the Diagnosis of Major Depressive...

Area of Science:

• Biochemistry
• Psychiatry
• Computational Biology

Background:

• Major depressive disorder (MDD) is a significant global health challenge and a leading cause of disability.
• Currently, there is a lack of validated biomarkers for the diagnosis and treatment of MDD.
• This necessitates the exploration of novel diagnostic and predictive tools.

Purpose of the Study:

• To investigate the diagnostic and predictive utility of serum amino acid concentration differences between MDD patients and healthy controls (HCs).
• To integrate these findings into interpretable machine learning models for potential clinical application.

Main Methods:

• Serum amino acid profiling was performed using chromatography-mass spectrometry on 70 MDD patients and 70 matched HCs.
• A total of 21 metabolites were analyzed, including 17 from an amino acid panel and 4 from a kynurenine panel.
• Logistic regression models were employed for classification and biomarker identification.

Main Results:

• The optimal machine learning model, incorporating feature selection and hyperparameter optimization, achieved an Area Under the Curve (AUC) of 0.76 for classification.
• Five key metabolites were identified as potential MDD biomarkers: 3-hydroxy-kynurenine, valine, kynurenine, glutamic acid, and xanthurenic acid.

Conclusions:

• Serum amino acid profiles hold promise as potential biomarkers for MDD.
• Interpretable machine learning models utilizing amino acid data can potentially enhance the diagnostic accuracy of MDD in clinical settings.