Physiological data were analyzed from only successfully completed trials, on average including 50 repetitions

for each combination of trial type (cues 1–3) and choice stimulus. All statistical analyses were performed using MATLAB (MathWorks). Unless otherwise specified, the instantaneous firing rate was estimated by the mean spike count within a 50 ms sliding window. The overall population response to each stimulus type was assessed relative to the prestimulus baseline period (−100 to 0 ms) with univariate statistical analyses. The activation state of the full population was represented as a 627-dimensional coordinate in Euclidean space, where each dimension represents the instantaneous firing rate of a single neuron estimated selleck compound within 50 ms sliding windows. The dynamic trajectory through this state space is the path that passes through the multidimensional coordinate of each time point. To explore the dynamic behavior of this network, we first calculated the Euclidean distance between trajectory pairs (e.g., D(Cue1,Cue2)) as a function of time. Distances between all contributing pairs of conditions (e.g., all pairs of cues) were then averaged to yield a single summary see more statistic for each time point. Randomized permutation tests were performed using exactly the same approach but with randomized

condition labels for each sample. Because distance is always positive, distance values were expressed relative to the median of the permutation test, and statistical significance was inferred relative to the observed null distribution. This distance measure is closely related to the population coding metric described below—accurate

pattern decoding critically depends on a reliable multidimensional distance between conditions. Position in state space Montelukast Sodium was visualized using classical MDS. For each analysis, we made four independent estimates of the population activity for each condition of interest by averaging every fourth trial within that condition. MDS was then performed on the Euclidean distance matrix. Data were plotted against the first two dimensions. The distribution of the four within-condition data points provides a sense of the tightness of condition-dependent clustering. We also consider the speed of activity trajectories through state space. In general, the instantaneous velocity at time t can be estimated by calculating the difference in activity state as a function of time: d(P1t-n, P1t+n)/2n. Here, we used n = 20 ms. As in all other analyses, firing rate was estimated within a 50 ms sliding window. This metric is equivalent to summing the absolute slope of each firing rate (see Kusunoki et al., 2009) and is sensitive to mean changes in the overall activation level of the system. However, most importantly, this metric is also able to identify changes in patterns that are not associated with a mean change in energy.