EEG Aging Manifold

This repository contains the manuscript, reproducible scripts, frozen model artifacts, and generated tables for a longitudinal resting-state EEG aging trajectory analysis.

Datasets

Development dataset:

• OpenNeuro ds005385, version 1.0.3
• DOI: https://doi.org/10.18112/openneuro.ds005385.v1.0.3
• Baseline model fitting uses session-1 eyes-closed pre-task EEG only.

External validation dataset:

• OpenNeuro ds003775, SRM Resting-state EEG, version 1.2.1
• DOI: https://doi.org/10.18112/openneuro.ds003775.v1.2.1
• Descriptor DOI: https://doi.org/10.1016/j.dib.2022.108647
• External validation uses first-session eyes-closed resting EEG in the adult subset age 20 years or older.

Raw EEG data are not committed. See data/README.md and scripts/00_download_or_link_data.md for download and placement instructions.

Repository Structure

• data/ - data-access notes plus ignored local raw/derived data locations.
• notebooks/ - exploratory notebook retained for provenance.
• scripts/ - numbered reproducibility pipeline.
• src/eeg_age/ - reusable feature, modeling, trajectory, stats, and plotting code.
• results/models/ - versioned frozen development model parameters.
• results/features/ - derived external feature table used for validation.
• results/projections/ - fixed-projection coordinates.
• results/tables/ - CSV tables regenerated from scripts.
• figures/ - regenerated external-validation figures.
• tables/ - LaTeX table fragments for the manuscript.
• manuscript/ - LaTeX manuscript and publication figures.
• outputs/ - existing ds005385-derived feature tables and legacy outputs.
• robustness_outputs/ - previously generated robustness analyses.

Environment

Using pip:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Using conda:

conda env create -f environment.yml
conda activate eeg-aging-manifold

Exact Run Order

Run from the repository root.

1. Download or link raw data as described in:

Get-Content scripts\00_download_or_link_data.md

2. If regenerating ds005385 features from raw EDF files:

python scripts\01_extract_features.py --dataset ds005385 --session ses-1
python scripts\01_extract_features.py --dataset ds005385 --session ses-2

The repository already includes the ds005385-derived feature tables under outputs/, so this step is optional unless starting from raw data.

3. Fit frozen development parameters from ds005385 session-1:

python scripts\02_train_development_models.py

4. Project ds005385 session-2 follow-up recordings without refitting:

python scripts\03_project_followup.py

5. Extract SRM external-validation features:

python scripts\01_extract_features.py --dataset ds003775 --force

6. Run external validation with frozen ds005385 parameters:

python scripts\04_external_validation.py

7. Regenerate figures and tables:

python scripts\05_make_figures.py
python scripts\06_make_tables.py

8. Run reproducibility sanity checks:

python scripts\07_sanity_checks.py

Frozen Model Artifacts

scripts/02_train_development_models.py saves the following under results/models/ds005385_ses1_v1/:

• feature_list.json
• feature_transform_params.csv
• pca_loadings.csv
• pca_params.json
• trajectory_params.json
• trajectory_age_calibration.json
• ridge_coefficients.csv
• ridge_params.json
• manifest.json

External validation loads these files directly. PCA, scaling, trajectory centroids, and ridge coefficients are never refit on session-2 or external data.

Current External Validation Summary

The independent SRM ds003775 adult subset contained 100 participants aged 20-71 years. The fixed-projection coordinate showed weak external linear association with chronological age (Pearson r=0.228, 95% bootstrap CI 0.010-0.414), but the rank association was not significant (Spearman rho=0.112, 95% CI -0.099 to 0.314, p=0.269). The transferred ridge brain-age model did not generalize as a calibrated predictor (MAE 31.36 years, negative R2).

Key Generated Tables

• tables/cohort_summary.tex
• results/tables/cohort_summary.csv
• tables/development_external_performance.tex
• results/tables/development_external_performance.csv
• results/tables/reproducibility_sanity_checks.csv