Dear BrainLat team, I am working with the BrainLat EEG release under a proof-of-concept contract and have two questions that currently block analysis of the MS arm. Both come from metadata inside the distributed .set files rather than from any modelling result, so I hope they are quick to settle. 1. What cohort does EEG_data/4_MS/AR contain? All 23 resting-state recordings in that directory record their own origin as a Parkinson’s preprocessing directory: comments: "Original file: /users/xbb23175/Documents/JCC/DataPreproc/PD/Suj_NNN.bdf" They also carry an empty group field, an empty subject field, and a generic setname: "BDF file resampled pruned with ICA" with no subject identifier. The 19 AR control recordings differ on every one of those: group: "CN", subject: "sub-NNNNN", and a subject-specific setname. The files are named suj_509–suj_535 on disk but reference s6_sub-1000NN internally, and those internal names are also real subject directories under 5_HC. The internal numbering appears to be a per-batch counter rather than a subject id — it runs sequentially in processing order and skips 100013 exactly where suj_523 is absent from the directory. I would therefore avoid joining on it, and would be grateful for confirmation that this is the intended reading. For reference, the PD release at the same site uses a disjoint identifier block (s401–s413), so these are not duplicates of those recordings. Could you confirm whether 4_MS/AR holds multiple sclerosis recordings, and if so what the authoritative subject identifiers are? 2. What preprocessing was applied to 4_MS compared with 5_HC? Two differences are measurable and both affect spectral analysis: Filtering. The control files record comments: "Step 0: filterAndResample = true ; newSR = 512 ; freqRange = [0.5, 40.0]". The 4_MS files record no filter parameters at all. If the two groups were band-limited differently, band-power comparisons between them are confounded at every frequency — relative power normalises by a sum across bands, so attenuating one band in one group shifts all of them. Amplitude. Reading the .fdt files directly, median absolute amplitude is 3.32 (range 1.66–6.39) for 4_MS and 0.76 (range 0.42–1.46) for 5_HC — fully disjoint, a factor of roughly 4.4. The reference field is average in both groups, and the two 4_MS files that record drift removal (suj_521, suj_532) are not lower-amplitude than their peers, so neither referencing nor an unmatched high-pass appears to explain it. The PD release files measure 5.19 (3.46–10.1), which overlaps 4_MS and sits well above 5_HC, suggesting the control pipeline applied a different scaling convention from the others. Could you tell us which gain or unit convention was used for each batch? The files were also written with different EEGLAB versions (2020.0 for 4_MS, 2022.1 for 5_HC) and retained different numbers of ICA components (≈125 vs ≈127). 3. Smaller items, if convenient Ages and sex for the 23 EEG recordings in 4_MS/AR. The demographic file gives ages for the MRI cohort, which we cannot join to these recordings. The EEG↔MRI subject mapping for MS: EEG_ID is empty for all MS rows in both metadata files. BrainLat_Demographic_MRI.csv and the cognition file map sub-AR00003 to different EEG_ID values. The Synapse entity for BrainLat_records_MRI (syn64707910) downloads as a Google Drive HTML page rather than data. Happy to share the metadata extraction script and its JSON output if that would help. Thank you for making the dataset available. Best regards, Augusto Capone

Created by Augusto Capone acapone13

BrainLat — cohort identity and preprocessing parameters for EEG_data/4_MS/AR page is loading…