Figure 3.

Mutual information analysis of PFC neural response. (a) MI between stimuli and pairs of PFC neurons depends on the size of the window of analysis and the time from stimulus onset. Each point in the heat map represents the averaged MI value across PFC neuron pairs for a given window size and a given time point measured from stimulus onset. MI values not significantly different than basal values (p > 0.05, Sign test) are displayed in black. Note that the averaged-across-time MI values (red line, <MI>) grows fast as the window size approaches 320 ms (dashed line), reaching 80% of the maximum <MI> computed across all window sizes. Further increments in window size does not result in higher <MI> values, for which W = 320 ms was chosen as the window size in subsequent analysis. (b) MI along the trial, computed in windows of 320 ms. After the onset of the auditory stimuli, MI reaches a peak at 250 ms and remains above chance from then on (mean ± s.e.m. is shown). (c) Linear decoders were trained to predict the tone identity (GO or NoGO) from the firing rate of pairs of PFC neurons or from their corresponding binary values. The plot shows the average classification performance computed across all pairs of PFC neurons. Note that performance for firing rates and binary values highly overlaps, suggesting that the binary model retains all information contained in the firing rate extractable by the linear decoders. Mean +/− s.e.m. of 100 bootstrapping iterations are shown. (d) Mutual information was higher for trials preceded by different tones (GO vs. NoGO, red line) than for trials preceded by the same stimulus (NoGO, blue line; *p < 0.01, Sign test). This implies that information conveyed by pairs of neurons in the PFC during tone presentation is primarily driven by stimulus-related events rather than the behavioural response itself (lick or no-lick).