Power estimates are not normally distributed across time samples, and thus we took the logarithm of power estimates in order to normalize their distributions (Miller et al., 2009). Prior to computing logarithms, each time course was divided by its mean value. This effective whitening of the high-frequency spectrum is not essential, but it slightly improved signal-to-noise in the estimate of high-frequency

power, because it corrects LY2157299 cost for the fact that lower frequencies exhibit larger fluctuations than higher frequencies. After whitening, one is combining spectral estimates across equally weighted independent samples of the underlying broadband process. Without whitening, the independent samples of the broadband process are not equally weighted. Broadband power was thus calculated as the average across all normalized time courses with center frequencies in the range 64–200 Hz. High-pass, low-pass, and band-pass filtering of power time courses Obeticholic Acid manufacturer (Figure 7) and voltage time courses (Figure 2) was performed directly in Fourier space, by computing a discrete fast Fourier transform (DFFT), separating the phase and amplitude of each Fourier component, multiplying the set of component amplitudes with the desired spectral profile, and

then inverting the DFFT. To attenuate time-domain ripples, a Gaussian taper was applied. For the 0.1 Hz cutoff, this taper produced 75% signal attenuation at 0.11 Hz, and >99% attenuation at 0.13 Hz. Comparable results were obtained using a time-domain Butterworth filter. The reliability of the power time courses evoked by each movie clips was then assessed using the Pearson correlation coefficient r=P1(t)×P2(t)‖P1(t)‖‖P2(t)‖,where P1(t) and P2(t) are time

courses of broadband power modulation evoked by the first and the second presentation of each clip. To avoid onset transients Mannose-binding protein-associated serine protease and horizon effects, the first 15 s and last 10 s of power modulation in response to each movie clip were excluded from all analyses. For analyses of the 30 s fixation periods, the first 5 s and last 5 s of each period were excluded. An audio amplitude time course was calculated separately for each soundtrack and then compared against the neural response time courses. Audio power modulations were estimated within 25 frequency bands (200 Hz to 5 kHz center frequencies, 200 Hz frequency width, 50 ms time width) using multi-tapers in FieldTrip. The logarithm was taken of the audio power time course in each band, and the “audio envelope” was computed as the mean across the audio power time courses in all bands. The audio envelope was then downsampled to the 10 Hz sampling rate of the neural power time courses. Finally, for each movie clip and each electrode, a Pearson correlation was computed between (1) the time course of the audio envelope, and (2) the average time course of broadband power for the first and second presentations of the clip.