Classification accuracy for various neuroimaging markers (50% of the data are used as training set and the rest 50% as test set, except for the last column, where a leave-one-out cross-validation is used).

The best results are achieved when we combine the features from the community matrix K and the asymmetry measure ρ. The accuracy of classification using SVM versus the number of edges selected from the community matrix K can be found in the Text S4.