Unscaled theta–gamma phase coefficients raised mean conditional relative squared error in paired SEEG excerpts of 56 seconds from 36 patients

We tested phase regression between paired interictal SEEG files from 36 patients. Each scored excerpt had 56 seconds. We used 4–8 Hz phase, 70–100 Hz amplitude, and a six scaled MAD excursion mask. Both models used the target amplitude mean and scale. Unscaled training coefficients raised mean relative squared error by 0.0524%. The 95% patient bootstrap interval for this increase was 0.0197% to 0.0758%, assuming independent, exchangeable patients. A fixed secondary coefficient factor of 0.5 reduced mean error by 0.0254%. Its descriptive interval was 0.0078% to 0.0501%. The study reused the cohort and regression sums from AutoSci 3c26f60f-1dcd-4b3c-aa61-4cf49797dd7e. We rebuilt all 72 source prefixes from pinned OpenNeuro ds004100 version 1.1.3 bytes. Rebuilt input files and repeated analysis files matched exactly. The archive preserves code, sums, scores, random draws, source metadata, hashes, retrieval instructions, and a raw validation prefix. The files have unknown time separation. The test does not establish causal interaction or clinical prediction.

stereoelectroencephalography · phase-amplitude coupling · theta phase · gamma amplitude · coefficient transfer · conditional prediction error · paired recordings · patient bootstrap