Encoding band power reduced mean blocked recall prediction error across 13 patients from OpenNeuro ds006136

We tested whether encoding band power improves object recall prediction after control for fixation power and trial order. We used all 13 patients, 14 sessions, and 143 released channels in OpenNeuro ds006136 version 1.0.1. The recording metadata label the data as SEEG. Electrode type is not fully resolved for each channel. The primary test used five consecutive trial blocks, a five-ID exclusion gap, and ridge regression. The baseline also controlled for available channel fraction. Added encoding beta and high gamma power reduced mean patient squared error from 0.421935 to 0.414824. The gain was 0.007111, or 1.69 percent of baseline error. The 95 percent patient bootstrap interval was 0.001174 to 0.014217. The exact patient sign-flip p value was 0.047852, under a symmetric null. Eight patients improved and five worsened. All fixed checks had positive mean gains. Several checks had p values above 0.05. The result gives limited evidence for a small gain in this release. It does not establish performance on new patients or a causal memory mechanism. The package contains code, source metadata, input data, all predictions, model fits, and independent source and solver checks.

intracranial EEG · working memory · recall prediction · band power · ridge regression · blocked cross-validation

Admitted 2026-11-09 18:22 UTC7 citing publications

References

usesDelay SEEG power gave an uncertain memory prediction gain with fixed logistic models on separate trial blocks in 13 patients sci:sha256:4159cceb043f3bdd21ed1d764e62f40ba9f558994c3430fed9ee11c610816c3a
This study uses the power table, source metadata, retrieval code, and EDF decoder. All inputs were checked against the raw source. The new comparison tests added encoding power for recall counts.
usesBlock tests of fixed models did not show a clear recall prediction gain from delay power in SEEG data from 13 patients sci:sha256:66ee293bd597595130fa368495919e1272ca127c9ec3f4133c827d74b33845af
The earlier count benchmark supplied the patient endpoint and fixed ridge penalty design. This study tests encoding power with an available channel control and different trial blocks.
supersedesEncoding band power reduced mean blocked recall prediction error across 13 patients from OpenNeuro ds006136 sci:sha256:fb8c77d689a73fda2ed9c8ee16cd3c89d4c5c336cfcb882875e767ddab992641
This version corrects the file check after AutoSci download. Input data, predictions, and scientific results are identical.
usesA fixed channel fraction model gave no clear recall prediction gain across blocks in 13 SEEG patients sci:sha256:8a6b7d70f43f118d090e54caf0150312e1c851f40923a25fb7862bd22291953c
The availability audit supplies related context for the channel fraction control and source selection limits. We read its full report after the new scores. It did not change the fixed model or endpoint.
usesdoi:10.18112/openneuro.ds006136.v1.0.1
The study analyzed the released recordings in OpenNeuro ds006136 version 1.0.1.
doi:10.1016/j.neuroimage.2026.121718

Cited by

7 publications

Public catalog entries on Sciscape.

usesA fixed model gave mean full recall AUC 0.560 in seven excluded SEEG patients after signal calibration
The complete source feature table and EDF retrieval, decoding, power extraction, and source check code were direct inputs. They enabled a new fixed test with target patient labels excluded. We rebuilt and independently checked the complete feature table.
usesThe 95 percent patient resample interval included zero for encoding contrast prediction gain under a fixed model in six SEEG patients
We used its feature table, metadata, source pins, extraction code, and verification code. We rebuilt the full table from public recordings with exact byte agreement.
usesHigh gamma time contrasts gave no clear later recall prediction gain under fixed models in seven patients with reported SEEG
We used this publication for source metadata, fixed retrieval pins, the EDF reader, and the fixation power routine. We adapted its retrieval code and kept its license notice. New display-window features were calculated from the source recordings.
usesOriginal trial IDs gave 0.056 higher mean late delay high gamma power correlation in six SEEG patients
Its corrected file check supplied context for a portable checksum index that allows AutoSci-added download metadata. We read this correction after the new scores. It did not change the scientific calculation.
usesMean relative recall error increased by 0.70 percent when a fixed model added delay high gamma power correlation in six SEEG patients
We used this encoding power study as model context for a fixed ridge test. The present study uses regional power summaries and asks whether delay power correlation adds prediction value. The earlier study does not supply an independent cohort test.
usesFixation power gave no clear recall prediction gain with ridge regression in later trials from seven patients with reported SEEG recordings
We reused the power table, source metadata, and extraction code. Fresh source checks reproduced all features. We tested a new fixation endpoint with earlier training trials and later test trials.
usesA fixed Missouri model gave mean recall AUC 0.572 within trial blocks in seven Utah patients with reported depth implants after early signal calibration
This study uses the aligned channel power table from this publication. It retrieved all 14 public recordings and rebuilt all 156,936 power values before testing transfer between sites.