Residual draws raised ordinal time asymmetry over Gaussian noise in AR(50) controls for two SEEG contacts in 36 HUP patients

We compared two noise models for linear SEEG controls. We used public interictal data from OpenNeuro ds004100, version 1.1.3. The data cover two selected contacts in each of 36 HUP patients. We fitted a linear model with 50 coefficients to 20–140 s in each source file. We measured three-point ordinal time asymmetry in a later 60 s window, with 20 ms spacing. Each fitted model generated 99 controls with Gaussian noise and 99 controls with independent draws from its training residuals. The mean residual-control score exceeded the Gaussian-control score by 6.55701e-05 nats. The 95% patient bootstrap interval was [4.27053e-05, 9.19535e-05]. The difference was positive in 34 of 36 patients. The difference remained positive at model orders 20 and 100 and at all four tested spacings. The measured SEEG score still exceeded the residual-control mean by 0.000160109 nats. Residual dependence and signal changes limit interpretation. This result shows that the noise model changes the linear baseline for this SEEG measure. It does not establish nonlinear brain dynamics. The package includes native traces, source pins, code, results, and audits.

SEEG · ordinal time asymmetry · time reversibility · linear controls · noise models · autoregressive model · Jensen-Shannon divergence · epilepsy