case, thresholding and binarization of the network matrix has been a typical approach to eliminate the weak connections. However, without an optimal threshold, the matrix binarization tends to overestimate the contrast-to-noise ratio of the network in that it enhances connectivity values above the threshold and hides the values below the threshold (van Wijk et al., 2010). We therefore chose the weighted network matrix instead of the binary one in the light of several studies showing that results from weighted networks are not different from the binarized matrix (Li et al., 2009; Ponten et al., 2009). Nonetheless, there is currently no gold-standard for selecting the best threshold for the matrix. In the present study, we applied a wide range of threshold values to each SL connectivity matrix, where the number of connections in the SL matrix was fixed according to the connection density (defined as the total number of connections in a matrix divided by the maximum number of connections) from 5 to 100% using increments of 5 (Fig. 5). For lower thresholds, the network tends to split in two or more subnetworks, resulting in the lower C and shorter L. As the threshold increases, both C and L also rapidly
