Detecting depression severity using weighted random forest and oxidative stress biomarkers

Area of Science:

• Computational psychiatry
• Biomarker discovery
• Machine learning applications in healthcare

Background:

• Major depressive disorder (MDD) diagnosis relies on subjective criteria.
• Objective biomarkers for MDD severity are needed to enhance diagnostic accuracy.
• Oxidative stress is implicated in the pathophysiology of depression.

Purpose of the Study:

• To develop and compare machine learning models for classifying MDD severity.
• To evaluate the contribution of oxidative stress biomarkers versus broader health factors.
• To improve diagnostic accuracy for major depressive disorder.

Main Methods:

• Binary and multiclass classification using machine learning algorithms (Random Forest).
• Analysis of data from 830 participants, including oxidative stress biomarkers, sociodemographic, and health factors.
• Application of techniques like weighted classifiers and Synthetic Minority Over-sampling Technique (SMOTE) to handle data imbalance.

Main Results:

• The Weighted Random Forest (WRF) model achieved an Area Under the Curve (AUC) of 0.91 in multiclass classification when all features were included.
• Including sociodemographic and health factors alongside oxidative stress biomarkers improved multiclass AUC from 0.84 to 0.88.
• Oxidative stress biomarkers, particularly Glutathione (GSH), were identified as highly important features.

Conclusions:

• Machine learning models, especially WRF, demonstrate high accuracy in detecting MDD severity.
• Integrating oxidative stress biomarkers significantly enhances the performance of diagnostic models.
• Clinicians can utilize these findings to improve MDD diagnosis by incorporating oxidative stress markers.