At each 0.1 Hz step, the time-series of instantaneous amplitudes φA at frequency FA and of instantaneous phase φP at frequency FP were extracted from real and imaginary time-frequency representations of the EEG segment to calculate raw PAC estimates for the phase-amplitude frequency pair (FP, FA) using the Phase-Locking Value (PLV) equationPLV(φP,φA)=1T|∑t=1Tei(φP(t)−φA(t)))|which measures consistency of phase differences between the phase φP and the phase of the amplitude envelope φA for all time points T in the signal (Lachaux et al., 1999; Seymour et al., 2017). A null distribution was generated by swapping amplitude time blocks prior to PAC estimation (200 permutations) to get expected mean value μ0 and standard deviation σ0 of PAC obtained by chance. Corrected PAC estimates were calculated by subtracting μ0 from the raw PAC estimates then dividing by σ0. Phase-amplitude comodulograms depicting the corrected PAC estimates across the phase frequency (x-axis) by amplitude frequency (y-axis) ranges were saved as grayscale images in JPG format and resized to 224 x 224 resolution (8-bit, 96 dpi) for use in a machine learning project described elsewhere (Richard et al.,
