In 8 ds004473 subjects, pre-fixation 70-150 Hz and 13-30 Hz SEEG power did not predict next-trial RT after drift removal (mean rho 0.009; 3/8 positive; p=0.383)

We tested whether archived adjacent-bipolar SEEG power in the 0.75 to 0.25 s before fixation predicted the next cued key press reaction time (RT) in 8 subjects from the OpenNeuro ds004473 SlowFast task (version 1.0.2). For trials with RT from 0.15 to 2.0 s, we used log RT and 70-150 Hz (high gamma) and 13-30 Hz (beta) features in ridge regression with 10 contiguous time-blocked folds. Linear trends over trial index were removed from features and log RT using training-fold coefficients. The primary group result showed no detectable prediction: mean Spearman rho was 0.009, 3/8 subjects had rho above zero, and the exact one-sided sign-flip p-value was 0.383. Neither band alone nor channel-mean features showed a detectable effect. Without drift removal, mean rho was -0.201 (1/8 positive; two-sided p = 0.023); this below-chance time-blocked result is consistent with a drift artifact and does not show an inverse relation. Foreperiod did not predict RT (mean rho -0.014, two-sided p = 0.648). One of 8 per-subject permutation tests was below 0.05 (sub-9, rho 0.14, p = 0.035; about 0.4 such results are expected by chance). The analysis was limited to 8 subjects in one task, a linear model, two bands, one rest window, and linear drift removal. The plan said the archived data contained the first 100 trials, but it contained all trials and all were used. An exact NumPy implementation replaced RidgeCV in the permutation loop because the host was slow; the primary rho values matched the RidgeCV checks exactly. The included plan, code, data, figure and results support reproduction.

SEEG · reaction time · band power · high gamma · beta · ridge regression · time-blocked cross-validation · drift · brain-computer interface · OpenNeuro ds004473