Get started
Work

Computational Code for An Explainable Multi-Modal Graph Learning Framework with Attention-based GCN-GAT for EEG-Based Depression Detection

Year: 2026
Type: software
Abstract: Frozen, leakage-controlled computational code supporting the manuscript “An Explainable Multi-Modal Graph Learning Framework with Attention-based GCN-GAT for EEG-Based Depression Detection”. Version 1.2.0 provides the subject-level population GCN–GAT scientific core, a reproducibility path for the canonical Combined EEG-only analysis (N=107; Figshare 56 + MODMA 51), Figshare-only EEG + XAI, and the multimodal analysis branches retained in the frozen scientific core. The release includes pooled outer-OOF aggregate metrics, strict-inductive and OOF integrity audits, fixed-topology Integrated Gradients, GradientSHAP, permutation feature importance summaries, aggregate reference figures, expected-result fingerprints, and package integrity metadata. Subject-level source data, the prepared analysis matrix, MODMA metadata, subject-level predictions, and subject-level attribution-long tables are not redistributed and must be regenerated locally after obtaining the source data under their original access and licensing terms. Performance values are internal cross-validated estimates, and XAI findings are model-conditional rather than independent biomarker validation. This software release supports a manuscript currently under peer review. Access to the deposited files is restricted during peer review and will be made publicly available following publication of the associated article. Editor access is provided through a private Zenodo sharing link.A separate supplementary archive, reviewer_validation_scripts.zip, has been added to provide scripts for reviewer-requested validation and sensitivity analyses. The original frozen v1.2.0 source-code archive remains unchanged.
DOI: 10.5281/zenodo.22107295
Language: English
Open Access status: green
OpenAlex ID: W7204274407
Cites:
Cited by:
Related to: 10