Figure 7
From: Brain Network Informed Subject Community Detection In Early-Onset Schizophrenia
Graphical demonstration of procedures to evaluate robustness to thresholds.
Two thresholds are used in the community detection algorithm: the probabilistic threshold, p, for binarizing the subject graph and the number of shared (overlapping) nodes between maximal-cliques, k, for merging the maximal-cliques into a subject community. For a given p, k values ranging from 2 to 6 are examined. For each choice of k, the community detection procedures are performed to generate a profile of subject membership, shown by the vectors indicating whether each subject is included in the subject community. Black color means a subject is in the community, while white color means a subject is not included. The cosine similarity between the profile vectors from different choices of k are computed to generated a similarity matrix. The sum along a given row indicates the reproducibility of that subject profile for a corresponding choice of k and the sum of the entire similarity matrix indicates the reproducibility of that subject profile for the given p. The p value with the maximal reproducibility value (sum of entire matrix) is first chosen and the optimal k is selected based on the maximal subject profile reproducibility (sum of row) under the given p.
