Fig. 3

Hierarchical community structure of protein networks. a, b Recovery of cellular components documented in GO by community detection methods (colored traces) versus number of top communities examined. Recovery is evaluated by the average F1-score. Communities are ranked in descending order of score for each community detection tool, similar to Fig. 2c (see the “Methods” section). A yeast network [23] and the human STRING network [27] were used as the inputs of a and b, respectively. HC.1–3 represent UPGMA Hierarchical Clustering of pairwise distances generated by Mashup, DSD, and deepNF [24,25,26], respectively. c Distributions of community sizes (x axis, number of proteins) for three human protein networks: BioPlex 2.0 [28], Coexpr-GEO [29], and STRING [27]. d Community hierarchies identified for BioPlex 2.0 (upper) or STRING (lower) databases. Vertex sizes and colors indicate the number of proteins in the corresponding communities. e Twenty-seven public databases of protein-protein interaction networks were analyzed by HiDeF and profiled by the maximum depths of their resulting hierarchies (y axis), which do not correlate with their total sizes (numbers of proteins, x axis; numbers of edges, color bar). f A hierarchy of communities of human proteins interacting with SARS-COV-2. The hierarchy, generated by HiDeF (see the “Methods” section), contains 252 communities of 1948 human proteins. Communities colored red are enriched (odds ratio > 1.5) for the 332 human proteins interacting with viral proteins of SARS-COV-2. Selected communities are labeled by gene set enrichment function provided in CDAPS (Availability of data and materials). g A community of interacting human proteins targeted by the SARS-COV-2 viral protein Nsp13 (see the “Methods” section). Direct interactors of Nsp13 (TLE1, TLE3, TLE5) are shown in orange