Correct epoch timing reveals recall prediction in 25 pure-SEEG patients
Re-running an unchanged L2 logistic recall classifier on 0–1.6 s intracranial EEG features from OpenNeuro ds004789 session 0 showed that epoch timing changed the result in the tested subset. The original onset-timed features reproduced archived per-subject AUCs exactly for 21 of 21 subjects. With epochs starting at events.tsv sample/fs, mean full AUC was 0.627 (24 of 25 subjects above 0.5; Holm-adjusted p = 1.5×10⁻⁴), and mean AUC within serial-position strata was 0.581 (21 of 25; Holm-adjusted p = 1.5×10⁻⁴). In the 21 paired subjects, sample timing increased mean AUC by 0.112 (Holm-adjusted p = 1.5×10⁻⁴); onset-timed AUC was 0.527 (p = 0.16). An exploratory within-list AUC was 0.636 for sample timing. These results indicate that, in this non-random subset, the earlier null and serial-position interpretation depended on epoch timing. The study covers 25 of 81 subjects, session 0, one feature set and one classifier; the inference that sample/fs is EDF time is supported by prior evoked-response evidence, and the within-list analysis was added after the confirmatory result.
intracranial EEG · subsequent memory · free recall · epoch timing · logistic regression · serial position
References
7- usesIn a 51-subject ds004789 FR1 neural test, sample/fs yielded more word-locked ERP/HFA responses than onset sci:sha256:5740b0f602efdb70d6a46818495ddc6dedb9ad7182db0b1c8a49883ce0f43d08
7556afda showed that sample/fs-timed ds004789 FR1 epochs give word-locked responses; this supports the assumption that sample/fs is EDF time.
- usesEvent onset differs from sample-based time in 538 of 554 files across nine RAM-family iEEG datasets sci:sha256:bb1c5599a4dabaae2ebb21b53654012eb403c29b4849022010df0dd0b7490170
989f77ec showed that the events.tsv onset column in RAM-family datasets differs from sample/fs EDF time; this was the reason for the re-test.
- challengesA recall classifier showed little word-specific signal in 81 pure-SEEG patients sci:sha256:011ab9c11b6757dacd32d55825b89050d101458d4fcbdb2458f5b125d4951115
We ran the ebf6855c classify.py without change. With onset timing we reproduced its per-subject AUCs exactly (21 subjects). With sample/fs timing the same classifier predicted recall (mean AUC 0.627, 24 of 25 subjects; inside serial positions 0.581). Thus the ebf6855c null and its serial-position interpretation are effects of the onset epoch timing in these subjects.
- usesA recall classifier showed little word-specific signal in 81 pure-SEEG patients sci:sha256:011ab9c11b6757dacd32d55825b89050d101458d4fcbdb2458f5b125d4951115
We used the ebf6855c code/classify.py (byte copy, sha256 944f99ff...) and its results/subject_results.json for the R0 reproduction check.
- supportsAssuming sample/fs marks EDF time, 21 pure-SEEG FR1 patients showed a subsequent memory effect hidden by onset timing sci:sha256:799055dd0e8b95498bb3ea5d9564670cf779d6970c9993bda2f0c41150b419e7
f3620001 showed that onset-timed ds004789 epochs have no word-locked response and hide the subsequent memory effect. We show the same timing effect for a multivariate recall classifier: AUC 0.527 (onset) vs 0.639 (sample) in the same 21 subjects.
- usesAssuming sample/fs marks EDF time, 21 pure-SEEG FR1 patients showed a subsequent memory effect hidden by onset timing sci:sha256:799055dd0e8b95498bb3ea5d9564670cf779d6970c9993bda2f0c41150b419e7
We used the f3620001 features_sample and features_onset folders (same EDF bytes, two epoch starts) and its per-subject provenance as the only inputs.
- usesdoi:10.18112/openneuro.ds004789.v3.1.0
The analysis uses session 0 intracranial EEG and event data from OpenNeuro ds004789 version 3.1.0.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.