EEG Multiverse

Note: The EEG data for this exercise is not publicly available. If you stumbled across this example outside of the workshop, please head to the solutions to inspect the results.

The data

Make sure you copy the data/ folder from the USB stick into the directory of this script.

from comet import multiverse

forking_paths = {
    "baseline":  [(-0.2, 0), (-0.1, 0)],
    "reference": [["A1", "A2"], "'average'"], # A1 + A2 are the mastoids
    "window":    [(0.400, 0.600),
                  (0.500, 0.700),
                  (0.450, 0.750),
                  "'SAV'"],
    "elec_set":  [["P3","P4","CP1","CP2"],
                  ["P3","Pz","P4"],
                  ["CP1","CP2"],
                  ["Pz"]]
}

def analysis_template():
    import mne
    import comet
    
    # Suppress warnings
    mne.set_log_level('ERROR')
    import warnings
    warnings.filterwarnings("ignore",module="pymatreader.utils")
    
    lpp_results = []
    subjects = ["21013", "21014", "21015", "21017", "21022"]
    
    for subject in subjects:
        # Load data
        raw = mne.io.read_raw_eeglab(f"data/eeg/{subject}_E4/{subject}_E4_postICA_interpolated_fixed.set", preload=True)
        
        # Epoch data
        event_map = {f"S {i:2d}": i for i in [18,19,28,29,38,39,48,49,56,58,68,69,78,79]}
        events, _ = mne.events_from_annotations(raw, event_id=event_map)
        epochs = mne.Epochs(raw, events, event_id=None, tmin=-0.2, tmax=0.8, baseline=None, preload=True, reject=dict(eeg=150e-6))
        
        # Baseline correction
        epochs.apply_baseline({{baseline}})
        
        # Re-reference
        epochs.set_eeg_reference({{reference}})
                
        # Crop (subject average peak or specified window)
        window = {{window}}
        if {{window}} == 'SAV':
            channel, time = epochs.average().get_peak(ch_type='eeg', tmin=0.3, mode='abs')
            epochs.crop(time-0.1, time+0.1)
        else:
            epochs.crop(window[0], window[1])
        
        # Pick electrodes for LPP estimation
        picks = mne.pick_channels(epochs.ch_names, include={{elec_set}}, exclude=[])
        
        # Get LPP as mean amplitude over epochs, channels & timepoints
        data = epochs.get_data() # (n_epochs, n_ch, n_times)
        lpp = data[:, picks, :].mean()
        lpp_results.append(lpp)
    
    # Save results
    comet.utils.save_universe_results({"LPP": lpp_results})

# Create and run the multiverse
mverse = multiverse.Multiverse(name="eeg_multiverse")
mverse.create(analysis_template, forking_paths)
mverse.run(parallel=8)
mverse.specification_curve("LPP", height_ratio=(1, 2), figsize=(10, 8), ci=95, baseline=0, title="LPP Analysis");

Starting multiverse analysis for all universes...

The multiverse analysis completed without any errors.
Warning: Only 5 samples were available for the t-test and CI.