A GRAPH-BASED BRAIN–GENE NOVEL FUSION MODEL FOR DEPRESSION CLASSIFICATION
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Depression is a complex neuropsychiatric disorder arising from interactions across molecular and neural systems, yet most computational approaches analyze neurophysiological and genetic data in isolation. This study proposes a graph neural network (GNN) – based framework for integrating electroencephalography (EEG) features and genetic markers to support automated depression classification. Using the multimodal subset of the ICBrainDB dataset, we analyzed 383 participants (349 controls and 34 individuals with depression) for whom both EEG and genetic data were available. EEG features derived from multiple scalp regions and resting-state conditions and gene-level variables from a curated panel were represented as nodes in a unified graph, with edges defined by train-only correlation structure. A GCN-style message-passing architecture with global pooling was trained using a leakage-controlled preprocessing pipeline and evaluated under class imbalance with comprehensive discrimination, agreement, and calibration metrics. On a held-out test set, the model achieved ROC-AUC = 0.9265 and accuracy = 0.9221, with high specificity (0.9857) and well-calibrated probabilistic outputs. Threshold optimization improved minority-class detection, yielding a best F1-score of 0.5455 and balanced accuracy of 0.7071. These results demonstrate that network-aware integration of EEG and genetic signals via GNNs can effectively capture cross-modal dependencies relevant to depression and provide a scalable, interpretable foundation for precision psychiatry applications.
Fields of Study & Subjects
EEG and Brain-Computer Interfaces
Emotion and Mood Recognition
Treatment of Major Depression
Publication details
- Journal / source
- Bezopasnost informacionnyh tehnology
- Volume
- 33
- Issue
- 2
- Pages
- 61–71
- Publication date
- 1 March 2026
- Language
- EN