We mapped the identified host proteins interacting with B. mallei onto a human PPI network [25] consisting of 76,043 physical PPIs among 11,688 proteins.

Of the 663 human proteins interacting with the nine B. mallei virulence factors, approximately 75% (498) were present in our human PPI network .

This set contained proteins that had, on average, a significantly larger number of interacting partners per protein (19.5 vs. 13.0) than would be expected from a corresponding random selection of proteins from the entire human PPI network (Table 3).

As the universe of human proteins, we used all constituent proteins from the human PPI network .

pestis PPI networks’ characteristics, see S7 Table.

Thus, we can exploit the fact that all three host-pathogen PPI networks contained interactions with human proteins that participated in similar biological processes.

Where deg(n) denotes the degree of node n and nd(n) denotes the neighborhood density of 11, defined as where N [n] = N(n)U{n} represents the closed neighborhood of node n, i.e., the node n and the set of its adjacent nodes (for a PPI network , this corresponds to a protein and all of its interacting partners).

For the topological node annotation, we used graphlet degree vectors [57] of all host and pathogen proteins from the host-pathogen PPI networks .

Topological properties of human proteins interacting with B. mal/ei in the human PPI network

We calculated the following topological properties for a set of human proteins interacting with B. mallei: 1) the number of human proteins interacting with B. mallei proteins (Np); 2) the average number of their interacting partners in the human PPI network (D); 3) the clustering coefficient, i.e., the number of interactions among the nearest neighbors (C); the average shortest path between any two proteins in the set (SP); the average number of interacting partners in the human PPI network where both partners interact with B. mallei proteins (Di); and the number of host proteins in the largest connected component (N itECG).

Multiple scoring schemes have been developed to infer protein functions in PPI networks , including algorithms based on network connectivity, graph topology and modular recognition [43—45].

Thus, the similarity between PPI networks and CSNs suggested that this approach could be effective for network-based drug target inference.

For example, it was shown that a Schwikowski score correctly predicted >70% of proteins with at least one functional category in a large-scale S. cerevisiae PPI network [43].

Likewise, the similarity between CSNAP networks and PPI networks provides further opportunities to apply different PPI network scoring schemes to improve CSNAP prediction [34].

For instance, neighbor counting functions could be readily expanded to consider second-order network neighbors, which has been shown to improve the prediction accuracy of PPI networks [67].

In Fig 6, enrichment was graphed in PPI networks as big yellow nodes for positive enrichment and small blue nodes for de-enrichment (defined as lower amounts in that fraction compared to elsewhere).

Nevertheless, PPI network analysis indicates that the phosphoproteomic data are complete enough to examine further to gain insight into signal transduction pathways that are active in neuroblastoma (S2 Fig).

(B) The most highly interconnected region of the neuroblastoma phos-phoproteomic PPI network (identified by the Cytoscape plugin, MCODE) is an almost perfect clique (a group where every node is connected to every other node).