Fig. 1.
Prioritizing network communities. a Community detection methods take as input a network and output a grouping of nodes into communities. Highlighted are five communities, (Ca, …, Ce), that are detected in the illustrative network. b After communities are detected, the goal of community prioritization is to identify communities that are most promising targets for follow-up investigations. Promising targets are communities that are most associated with external network functions, such as cellular functions in protein–protein interaction networks, or cell types in cell–cell similarity networks. CRank is a community prioritization approach that ranks the detected communities using only information captured by the network structure and does not require any external data about the nodes or edges of the network. However, when external information about communities is available, CRank can make advantage of it to further improve performance (Supplementary Notes 9 and 10). CRank starts by evaluating four different structural features of each community: the overall likelihood of the edges in the community (likelihood), internal connectivity (density), external connectivity (boundary), and relationship with the rest of the network (allegiance). CRank can also integrate any number of additional user-defined metrics into the prioritization without any further changes to the method. c CRank then applies a rank aggregation method to combine the metrics and d produce the final ranking of communities