20-second theta phase fits increased mean normalized amplitude prediction error across paired SEEG files in 35 patients under fixed rules

We tested transfer of theta phase fits with public SEEG data from OpenNeuro ds004100, version 1.1.3. We used 3,412 bipolar pairs from 35 patients, with two files per patient. Each fit and test used 20 seconds. Cosine and sine of phase were centered in each sample. Amplitude was centered and scaled by its own sample standard deviation. The mean across-file score was -0.1175% (95% patient bootstrap interval -0.1348% to -0.0951%). Zero means the same squared error as the zero-prediction baseline. Within-file transfer gave -0.0963%. Across-file transfer was worse by 0.0212 percentage points (95% interval 0.0112 to 0.0316). Fixed amplitude and pair weight checks also gave negative mean scores. These short fits imposed a small prediction cost under the tested rules. This result does not establish absent neural coupling or clinical value. This secondary study used previously inspected patients. The package preserves moments, code, raw examples, source versions, results, and repeat checks.

SEEG · phase-amplitude coupling · theta phase · transfer · prediction error · epilepsy · signal processing