Examining individual variation in brain age estimates during typical development.
Python 3.6.10
Required packages include: numpy, scipy, shap, scikit-learn, nibabel
All installed packages are shown in req.txt
To clone environment try: conda create -n new environment --file req.txt
PING data is available from the NIMH Data Archive subject to data use agreement. Study link: https://nda.nih.gov/study.html?id=905 (requires NDAR login)
1. run_brain_age_models.py
Load surface data, parcellate and preprocess then train and test brain age models and calculate individual model explanations
Output:
- Cross-validated model accuracies
- Age predictions
- Model explanations
2. run_variance_partition.py
Estimate variance explained in brain age delta by confounding variables
Output:
- % variance explained in delta by confounds and explanations
3. run_deconfound_data.py
For each model, remove variance associated with confounding variables from delta and model explanations using linear regression (performed within 5-fold cross-validation folds)
Output:
- Model explanations and brain age delta estimates with variance due to confounds removed
4. run_surrogates.py
Use BrainSMASH to generate random surrogate maps with matched spatial autocorrelation
Output:
- Surrogate maps (n_subjects x p_features x s surrogates)
5. run_explanation_correlations.py
Measure mean similarity of model explanations within subjects (across train/test folds), mean similarity between each subject and every other and mean similarity between each subject and set of random surrogates
Output:
- Mean cosine similarity for 'within', 'between' and 'random' comparisons