We then proceeded as for source-level coherence, but without neighborhood filtering. This resulted in clusters that represent significant changes in signal power across space, time, and frequency. We compared conditions using both random effects (across

subjects) Cilengitide and fixed effects (pooled across subjects) statistics. To visualize the identified networks we separately projected them onto different subspaces. To display the spatial extent (Figures 3A and 4A), we computed for each location the integral of the corresponding cluster in the connection space over time, frequency, and target locations. This integral was then displayed on the brain surface. This visualization reveals the spatial extent of the network independent of its intrinsic synchronization structure and location in time and frequency. Complementary to the spatial projection, we visualized the spectro-temporal projection (Figures 3B and 4B) by integration over all spatial locations (3D × 3D). This projection shows when and at which frequencies a cluster was active irrespective of the spatial location of synchronization. To analyze further properties of a network (modulations in power, other coherence contrasts, and single-trial analysis), we proceeded as follows: To account for interindividual differences, for each subject, we identified the connections within the network that were statistically significant (we computed

t-statistics for each connection in the cluster between conditions using STCP;

p < 0.05, one tailed). We averaged the property www.selleckchem.com/screening/anti-cancer-compound-library.html of interest (e.g., signal power) across each subject’s significant connections and used the resulting values for further analyses and tests. Importantly, the statistical sensitivity of these secondary tests is much higher than for the initial network-identification. The network-identification accounts for a massive multiple-comparison problem, whereas the secondary analyses use only a single test. This explains why the beta network differs between bounce and pass trials, as shown by a secondary analysis, but is not identified in the less sensitive network identification based on the bounce versus Bumetanide pass contrast. To analyze the synchronization pattern of the beta network (Figures 3C and 3E), we defined seven regions of interest (ROIs) in source space (Table S1). We selected sources that constitute a local maximum in the spatial network pattern and summed the connections between any two ROIs in the network. For each connection between two ROIs, the result was normalized by the maximum across all ROI-pairs, thresholded at 0.1, and visualized as the width of lines connecting the ROIs on the brain surface. We used ROC analysis to test whether coherence within a network predicted the subjects’ percept on a single-trial level (Green and Swets, 1966). We computed a predictive index that approximates the probability with which an ideal observer can predict the percept from the coherence on a single trial.