FIG. 5. Results of non-negative matrix factorization.

(a) Ranked differences in participation coefficients for all brain regions and with the number of communities ranging from 2 to 150. This matrix served as input to the NMF algorithm. (b) NMF identified four factors that approximately reconstructed the original data. Broadly, these factors were associated with different topological regimes corresponding to different numbers of communities. In panels c-f we show the factors mapped onto the cortical surface. Note that the first three factors emphasize participation coefficient differences in motor and visual systems while the fourth factor highlights widespread, multi-system differences. (g) We show system-averaged and z-scored differences in participation coefficient. Note that the first three factors all emphasize motor and visual systems but in different proportions. (h) Scatterplot of first two factors against one another. Note that somatomotor and visual areas tend to display the largest values in general. However, note also that those areas deviate from the diagonal, indicating that differences in these systems are emphasized differentially by the two factors and at different topological scales.