Fig. 1
From: Robust, scalable, and informative clustering for diverse biological networks
Overview of the clustering process in SpeakEasy 2: Champagne. A Each node in the network receives a random label, with the total number of labels less than the total number of nodes. B Each node updates its label to the most unexpectedly common label among its neighbors, accounting for the global frequency of each label. C Large clusters may mask multiple true communities. D Large ill-fitting clusters are split into random labels. E Stable sub-cluster configurations may occur which are not globally optimal. F By operating at the level of complete modules, suboptimal clusters can be split or merged to find globally optimal clustering states. G Multi-community nodes are identified based on those nodes which typically join a small number of distinct clusters, across multiple independent runs of SE2. H Overarching sequence of stages in SE2 algorithm, individually described in prior panels