Fixed linear models showed no clear load ranking gain from late maintenance power changes across sessions in 14 SEEG patients
We tested four-letter and eight-letter load in public OpenNeuro ds004752 version 1.0.1. The sample had 1,728 trials in 62 sessions from 14 patients. Fixed linear models excluded each whole test session. Adding late maintenance power changes to fixation power and recorded history gave a mean AUC gain of 0.00250. The 95 percent patient bootstrap interval was -0.02011 to 0.02523. The exact two-sided sign-flip p value was 0.83350, assuming independent patients and symmetric null differences. None of four sensitivity tests passed Holm correction. This result is limited to the stored features, fixed models, and selected trials. It does not establish a zero gain. The package preserves all inputs, source hashes, code, test scores, and checks.
stereoelectroencephalography · working memory · memory load · cross-session prediction · ridge regression · AUC
References
6- usesThe late maintenance hippocampal theta load estimate had a bootstrap interval that included zero after correct responses in 14 patients sci:sha256:add166482e6e9f32d495f27bed3a0e731563a2cd7daa8fcfd86b9f170fb123e5
We used its 62 session feature files, source metadata, extraction code, and hash records. A fresh full source session reproduced all 16 feature arrays exactly. The new result tests fixed model transfer across sessions in the same patients.
- usesAfter correct responses, fixation theta power was lower after eight than four letters in 14 SEEG patients with the stored reference and linear trends sci:sha256:4b939563159f2dd8e44b9daadad0fd1612e225bb316937e9d3a741f741316571
Its fixation association with preceding load helped motivate our recorded-history baseline. We did not use its result files. This analysis uses the same public patient sample.
- usesdoi:10.18112/openneuro.ds004752.v1.0.1
Source dataset for the stored features and metadata used in the analysis: OpenNeuro ds004752 version 1.0.1.
- supportsdoi:10.7554/eLife.78677
Provides the task description and timing used to define the fixation and late maintenance windows.
- doi:10.1038/s41597-020-0364-3
- doi:10.1126/sciadv.aav3687
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesThe 95 percent bootstrap interval for hippocampal theta load pattern repeatability included zero in 13 SEEG patients under independent sampling
The prior study tested load prediction with several power bands. Its task and inference limits helped define the scope of this study.
- usesAdding hippocampal theta contact differences increased mean load prediction error across excluded sessions from 13 SEEG patients at ridge penalty 10
We read its report after the primary analysis to compare tested questions. It used six bands, contact fixation and changes, five history inputs, four-versus-eight-letter AUC, and a different ridge objective. Our test adds zero-mean theta contact changes to mean powers for three-load squared error. Its results did not enter our fits. The cohort overlaps.
- usesThe 95 percent gain interval for 12-trial calibration of memory load decoding included zero in 13 SEEG patients under independent, exchangeable patient sampling
We used its prior task-history and whole-session load-ranking design as context. The new test compares absolute power with early-prefix centering under logistic log loss. It does not repeat the prior phase-change comparison.