Decoding depression: Event related potential dynamics and predictive neural signatures of depression severity

Area of Science:

• Neuroscience
• Psychiatry
• Computational Biology

Background:

• Depression diagnosis relies on subjective self-reports and interviews.
• Objective neurophysiological markers could improve depression assessment accuracy.
• Major depressive disorder (MDD) involves cognitive and affective processing disruptions.

Purpose of the Study:

• To investigate the utility of event-related potentials (ERPs) in classifying MDD and predicting depression severity.
• To determine if machine learning models trained on EEG data can differentiate MDD from healthy controls (HCs).
• To assess the potential of ERPs as objective biomarkers for depression diagnosis and severity stratification.

Main Methods:

• Participants underwent electroencephalography (EEG) while reading scenarios varying in predictability and affective valence.
• ERPs time-locked to critical words were analyzed, focusing on Late Frontal Positivity (LFP), N400, and Late Posterior Positivity (LPP) components.
• Machine learning classifiers were trained to predict clinical diagnosis (MDD vs. HCs) and depression risk using validated scales (BDI-II, PHQ-9).

Main Results:

• Machine learning models achieved 80% accuracy in distinguishing MDD from HCs.
• ERP features reliably identified individuals at high risk for depression based on self-report scales.
• Late Posterior Positivity (LPP) was most predictive of diagnosis, while N400 and LFP predicted symptom severity.

Conclusions:

• ERPs derived from EEG can serve as objective biomarkers for depression.
• Distinct neurocognitive processes are associated with diagnostic classification versus symptom severity.
• This machine learning-based neurophysiological approach supports data-driven, personalized psychiatric evaluation for depression.