Different Algorithms (Might) Uncover Different Patterns: A Brain-Age Prediction Case Study

Code for the paper "Different Algorithms (Might) Uncover Different Patterns" first presented in the 2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM 2023) [ArXiv].

Overview:

Machine learning techniques are ubiquitous in bioinformatics and EEG research in particular. However, there is still a lot of ambiguity and room for experimental design choices in the experiment setup and analysis pipeline. This raises an important question: Do different experimental design choices, particularly regarding model selection, influence the patterns uncovered in EEG data? Such variations could potentially result in differing, or even contradictory, interpretations.

Preprocessing Framework

Feature Selection & extraction

Model Training & Tuning

Performance On The NeuroTechX BrainAge Prediction Dataset

Computing The SHAPAgreement Across Models

data: we have provided sample data we used in the paper for demonstration purposes.
- example_eeg: contains eeg from one subject including eyes opened and closed states.
  - example_subj_EC_raw.fif.gz: sample eeg recording for eyes closed resting state.
  - example_subj_EO_raw.fif.gz: sample eeg recording for eyes opened resting state.
  - example_subj_EC_preproc.pickle: preprocessed eeg data from example_subj_EC_raw.fif.gz.
  - example_subj_EO_preproc.pickle: preprocessed eeg data from example_subj_EO_raw.fif.gz.
- example_training_set: contains the 12-all training set we used in the paper.
models: contains the corresponding code for each of the 10 models explored in the original paper.
- training.ipynb: contains the code for training and tuning the models.
- shap_values.ipynb: contains the code for computing the shap values.
- shap_values.pickle: is a dict object containing the shap values and fold indexes they were computed on for the specific model.
example_eeg_preprocessing.ipynb: a step by step notebook showcasing how to to run our data prerpocessing pipeline.
example_eeg_feature_extraction.ipynb: a step by step notebook showcasing how to run our feature extraction pipeline.
example_plot_interpolation_map.ipynb: a step by step notebook showcasing how to create the regional interpolation plots.
example_aggregated_shap_value.ipynb: a step by step notebook showcasing how to aggregate shap values and create rank orders of the feature groups based on the shap value.
example_shap_agreement_metric.ipynb: a step by step notebook showcasing how to compute the ShapAgreement metric between models.

Cite

@INPROCEEDINGS{10385662,
  author={Ettling, Tobias and Saba-Sadiya, Sari and Roig, Gemma},
  booktitle={2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)}, 
  title={Different Algorithms (Might) Uncover Different Patterns: A Brain-Age Prediction Case Study}, 
  year={2023},
  pages={4051-4058},
  keywords={Machine learning algorithms;Focusing;Prediction algorithms;Brain modeling;Feature extraction;Robustness;Electroencephalography},
  doi={10.1109/BIBM58861.2023.10385662}
}

Arsu-Lab / Different-Algorithms-Uncover-Different-Patterns-BrainAge-Prediction