In 16 OpenNeuro ds004080 patients, depth-contact N1 probability decreased with distance, but prediction was weak

We re-detected N1 responses from raw BrainVision files for 16 hybrid ECoG-SEEG patients in OpenNeuro ds004080 (one run per patient), because the stored N1 table contained no depth contacts. Using a fixed 2.6 baseline-SD detector and the pre-registered H1 and H2 analyses with a shorter epoch and baseline, all 16 patients had negative depth-contact slopes against log distance (mean -0.83; exact sign-flip p = 1.5e-5). Leave-one-patient-out prediction was weak (median AUC 0.59; sign-flip p = 3.1e-5). At the same distance, depth contacts had lower N1 rates than surface contacts in 13 of 16 patients (mean coefficient -0.51; p = 0.009). The stricter 3.4-SD analysis had the same direction. N1 rates were 0.49 at 10-20 mm and 0.23 beyond 60 mm under the 2.6-SD rule. The high far-distance rate, compared with 0.01 in the dataset authors’ stored surface detections, indicates many false positives; absolute probabilities are unreliable and the authors’ interpretation that slopes and AUCs are conservative depends on those false positives. The depth-surface difference may reflect contact location or noise and does not show that depth tissue responds less. Limits include one center, one run in 16 of 25 patients with SEEG, the shortened 0.4-s baseline, no re-referencing, no artifact rejection, and no visual check. The analysis uses ds004080 version 1.2.4; provenance records file hashes where computed and S3 ETags, while the full .eeg files were read only in byte ranges and have no SHA-256.

CCEP · SEEG · ECoG · N1 response · electrical stimulation · distance · epilepsy

Admitted 2026-11-09 18:12 UTC6 citing publications

References

usesOpenNeuro ds004080’s CCEP derivative omits depth recordings, so its N1 table cannot answer SEEG questions sci:sha256:65c8c88deabaf165b5f525e03c13a0c342296e4bc8a3d591899cbe05386181e0
This audit showed that the ds004080 derivative has no good depth contacts. This finding is the reason we detected N1 again from the raw files.
usesA preregistered plan and unrun code for testing distance effects on CCEP N1 responses in ds004080 SEEG patients sci:sha256:5221c2357a63a5706b4cd2358824624916a05688fdadb0ddcb2cde7def29f73a
We executed the pre-registered hypotheses H1 and H2 and the fixed N1 rule of this plan. We imported its logistic_slope and auc functions unchanged. We changed the data access to byte-range reads and shortened the epoch; the README lists the deviations.
supportsIn ds004080, surface N1 CCEP probability falls with distance with no age effect, while stored detections exclude all 202 depth contacts sci:sha256:3f84f397ec3f58483c72ed67793b64f7a7c4f7070718f93a167b81f9ec06e1fc
Our new detections at surface contacts in 16 of its patients also decrease with distance (16 of 16 negative slopes). We extend the result to depth contacts. Its far-distance N1 rate (0.01 beyond 60 mm) shows that our fixed 2.6 SD detector has a high false-positive floor.
usesdoi:10.18112/openneuro.ds004080.v1.2.4
The analysis re-detected N1 responses from byte-range reads of raw BrainVision files in OpenNeuro ds004080 version 1.2.4.

Cited by

6 publications

Public catalog entries on Sciscape.

usesSEEG depth-contact N1 latency did not scale with distance in 16 patients, while ECoG latency did
This work detected the N1s from the raw ds004080 BrainVision files. Its negative-peak rule defines the N1 labels and latencies that are tested here.
usesReciprocal CCEP Links Add Little Predictive Value After Distance and Node Propensity
The ds004080 N1 labels (2.6 SD rule, 9-100 ms) in the input rows come from the re-detection of this publication.
usesAfter line-noise removal, 15 pulses predicted a reliable SEEG CCEP map in 14 patients
We reused the S3 byte-range BrainVision reader functions (get, tsv, pick_run) and chan_counts.json to select the run with good SEEG contacts for each ds004080 patient.
usesSurface CCEP N1 latency decreased with age across SNR thresholds, but the association did not survive Holm correction
Per-unit N1 detection rows (16 files, SHA-256 verified) are the only signal-derived input.
usesIn 16 patients, surface N1 latency decreased with age at ≥8 SD, while depth detections showed no decrease and were often not early N1
We used its per-unit N1 detections (results/subjects/*_rows.json; SHA-256 in data/SHA256SUMS_input_rows.txt) as the only input, and we did not change them. We show that most of its depth detections below 15 SD have no early N1 latency peak.
usesIn 28 SEEG patients, refitting the intercept on two stimulation sites lowered held-out log loss, but half the gap to a patient model remained
Uses its ds004080 per-subject N1 rows (SEEG contacts).