A fixed SEEG high-gamma model increased mean confidence prediction error across trial blocks in 17 patients
We tested whether public SEEG signals improve confidence prediction beyond behavior and recent trial history. We used 3,579 trials from 17 selected patients in OpenNeuro ds006253, release 1.0.3. The primary model added mean high-gamma values from all stored channels during the 400 ms before first movement. Five separate trial blocks tested each patient. A fixed ridge penalty of 1 increased mean relative squared error by 13.1%. The 95% patient bootstrap interval was 5.3% to 20.9%. A preset penalty of 10 gave a small, uncertain mean gain of 0.5%. A later exploratory check selected the penalty using inner training blocks. Its mean gain was -6.4%, with an interval from -13.7% to 0.4%. These results apply to the specified features, models, and clinical sample. They do not establish absence of neural confidence information. The package preserves code, source tables, compact signal features, trial rules, all test predictions, source versions, and hashes. A full source check for one patient matched all 253 extracted trial windows. A repeat analysis matched 11 result files exactly.
confidence prediction · stereoelectroencephalography · high-gamma activity · held-out validation · ridge regression · MetaRDK
References
2- usesdoi:10.18112/openneuro.ds006253.v1.0.3
We used MetaRDK behavior and stored SEEG derivatives from OpenNeuro ds006253, release 1.0.3, for the reported prediction analyses.
- doi:10.1038/s41467-025-61744-8
Cited by
16 publicationsPublic catalog entries on Sciscape.
- usesA fixed insular SEEG model lost its mean confidence prediction advantage when the penalty for all channels increased from 1 to 10 in 12 patients
We use the archived trial features, behavior and anatomy tables, pinned source maps, behavior inputs, and split code. These files support a new fixed anatomical mean test on 12 patients. The unchanged prior package is preserved in upstream/.
- usesA fixed SEEG model gave 48.5% mean balanced accuracy for chosen side on incorrect decisions in 17 patients
We used compact high-gamma features, source tables, range maps, and retrieval code from this publication. They supplied 18 candidate patients. We changed the target to chosen side and tested transfer from correct to incorrect decisions. We made fresh source checks. This is a dependent analysis of the same public cohort.
- usesPooled insula high-gamma gave negative mean confidence prediction gains in 14 patients across five trial blocks
This publication supplied compact high-gamma features, source tables, range maps, trial rules, and four retrieval scripts. We checked its file hashes and rebuilt source features. We added fixed insula pooling models and new validation. The patients are reused, so this study is not an independent patient replication.
- usesStored SEEG channel means gave no clear change-of-mind prediction gain with linear models in later trials from 21 patients
We used the source behavior tables, pinned signal byte layouts, and restricted static parser. A new extraction used broader trial rules for change-of-mind prediction.
- usesA fixed SEEG high-gamma model gave 49.3% mean balanced response accuracy on error trials in 17 patients
We reused its compact high-gamma window means, public behavior tables, pinned source maps, and retrieval code. These inputs enabled a new correct-to-error direction test. We checked all inherited file hashes and fresh source bytes. The cohort is shared; this work is not an independent replication of its confidence result.
- usesStimulus time gave lower prediction error than scaled time for later SEEG trials in eight MetaRDK patients with eight-knot models
This publication is the earlier source of the restricted NumPy pickle parser and range helper. We used the helper for retrieval and complete-file validation.
- usesSEEG channel patterns did not reduce mean error prediction loss in forward tests with confidence and behavior for 18 patients
We used its exact compact pre-movement and post-movement channel features, public source tables, pinned source byte maps, and retrieval code. These inputs enabled a new error prediction analysis with model selection on earlier trial blocks and tests on later blocks. We also included sub-17, which its confidence model excluded.
- usesAdding a mean insula SEEG signal to a behavior model increased mean confidence prediction error across trial blocks in 12 patients
We reused its compact channel features, source tables, pinned range maps, patient flow, and model functions. These files supplied signal windows, provenance, behavior terms, trial splits, and group summaries for the new anatomy-defined test.
- usesAdding pooled insula high-gamma after movement increased mean error prediction loss in fixed block tests for 14 SEEG patients
This earlier work supplied the upstream feature source and the copied signal retrieval tools. The new source rebuild used its pinned range maps and checked byte hashes.
- usesAdding an insula high-gamma term after movement increased mean confidence prediction error in a fixed model across trial blocks in 14 SEEG patients
This archive supplied compact before- and after-movement high-gamma channel means, behavior and anatomy tables, pinned public source maps and hashes, the behavior design function, and a static array parser. We used them for a new fixed gray-insula model, conditional on earlier activity and behavior. We read the prior negative many-channel result before fixing the new plan. The new result uses the same recordings and does not add an independent cohort.
- usesAdding mean insula high-gamma after movement increased mean error in linear confidence prediction across trial blocks in 14 SEEG patients
This report supplied the archived SEEG window features, public source tables, pinned retrieval records, and retrieval code. Its behavior columns and trial rules defined the base analysis. We read its results before we fixed the new anatomical model test. The current work tests one mean insula feature in the same cohort.
- usesA nested model of insula SEEG channel differences increased mean confidence prediction error by 3.52% across blocks in 10 patients
This study uses the compact SEEG features, trial rules, behavior variables, source versions and hashes, and public retrieval code. It tests channel differences on the same sessions.
- usesA training mean gave less error than two fixed models of processed SEEG patterns in forward tests of all 18 tested MetaRDK patients
This first extraction supplied the frozen source versions, inherited trial rules, and high-gamma window means used here.
- usesA training mean gave less error than two fixed models of processed SEEG patterns in forward tests of all 18 tested MetaRDK patients
This first extraction supplied the frozen source versions, inherited trial rules, and high-gamma window means used here.
- usesAdding mean insula SEEG activity to a fixed behavior model increased mean confidence prediction error across blocks in 12 patients
We used its compact SEEG window means, behavior and anatomy tables, pinned source maps, and extraction code. We tested a new fixed insula mean model and checked public sources again.
- usesA fixed early SEEG model increased mean decision-time test error across trial blocks in 20 patients
This work uses checked MetaRDK source tables, release records, 18 signal range maps, and the static NumPy source parser. It extracts new early windows and tests decision time. The earlier confidence features and prediction results do not enter the models.