Code for: "ComBat models for harmonization of resting-state EEG features in multisite studies"

Benchmarking Combining Batches (ComBat) models on spectral parameterized features of resting-state EEG signals.

• The original sources/links to raw rsEEG data are available in our manuscript.

  • To download LEMON dataset, we recommend the following steps:
  • First, you will need to download and unzip the LEMON behavioral and demographics file (META_File_IDs_Age_Gender_Education_Drug_Smoke_SKID_LEMON.csv) available at the LEMON website.
  • As the original names in LEMON have changed, the LEMON website suggests to use these new participant_ids.
  • Run download_data_lemon.py modified from Engemann D, et al. NIMG 2022, available here.
  • Finally, convert LEMON to BIDS.

• Automated preprocessing was performed using sova-harmony in Python.

• Spectral parameterization was made with Fitting Oscillations and One-over Frequency (FOOOF - now specparams).

• Benchmarked ComBat models comprised:

  • neuroCombat (standard ComBat).
  • neuroHarmonize (variant with nonlinear adjustment of covariates).
  • OPNested-GMM (variant based on Gaussian Mixture Models to fit bimodal feature distributions).
  • HarmonizR (variant based on resampling to handle missing feature values).

• The full code to replicate figures and results is implemented in Python (combat_harmonization_eeg.ipynb) and R (harmonization_figures.R)

• The tsne folder has the .csv files required to generate Figures 3 - 5 (look harmonization_figures.R).

Qualitative visualization of batch effects

Statistical testing across batches

Downstream analysis (age-related spectral changes)

Cite as: Jaramillo-Jimenez, A., Tovar-Rios, D. A., Mantilla-Ramos, Y.-J., Ochoa-Gomez, J.-F., Bonanni, L., & Brønnick, K. (2024). ComBat models for harmonization of resting-state EEG features in multisite studies. Clinical Neurophysiology. 2024. https://doi.org/10.1016/j.clinph.2024.09.019