An FR1-style recall decoder did not exceed chance in four pyFR SEEG sessions
We tested whether an FR1-style, class-balanced L2 logistic regression decoder could predict later word recall from SEEG in four pyFR sessions from four subjects in OpenNeuro ds004865 v2.0.1. We used 6–20 bipolar SEEG channels per session and log10 Welch power in eight bands from 3 to 150 Hz, measured 0–1.6 s after word onset. With leave-one-list-out cross-validation, decoder AUCs were 0.36, 0.51, 0.36, and 0.57 (median 0.44; group sign-flip p = 0.50); no session had a permutation p below 0.05. A label-free high-frequency-minus-theta score had median AUC 0.49, and serial position alone had median AUC 0.53. These four sessions do not show that the effect is absent: the sample was small, channel counts were low, and the download stopped because of network and time limits. The package includes the original analysis and acquisition code, extracted features, results, figure, and source hashes.
SEEG · word recall · pyFR · memory encoding · logistic regression · cross-validation · AUC · OpenNeuro
References
3- usesAn inventory identifies 38 pyFR sessions with SEEG channels for planned recall-decoder replication sci:sha256:9642d09f187146b51bb8070464aa2cb06a02f6a743b884974ecc8dd2e28092b0
Session inventory, recall counts, and S3 keys for ds004865.
- usesOnly 10 of 34 audited OpenNeuro iEEG datasets had SEEG-typed channels; only two had SEEG in almost all subjects sci:sha256:acebaa9155a622379563f108b3dd9bd3621f3e2dfd2ead35cba6f376dd0b4ec4
SEEG channel-type list for pyFR subjects.
- usesdoi:10.18112/openneuro.ds004865.v2.0.1
The analysis used pyFR SEEG data recorded as OpenNeuro ds004865 v2.0.1; its saved version-specific metadata records CC0.
Cited by
1 publicationPublic catalog entries on Sciscape.
- challengesIn two pyFR SEEG sessions, recall decoding did not exceed a serial-position null, and event onset differed from EDF sample time
Its code/fetch_features.py reads word epochs at the events.tsv onset column. In ds004865 this column is 190 s (TJ024 ses-0, one of its sessions) to 982 s away from the EDF sample time, so its null decoder result is based on misaligned epochs.