Fig. 3

Network visualization reveals network modularity and increased self-loops in epithelial-stromal co-expression networks in invasive breast cancer. The epithelial-stromal co-expression networks are visualized for a normal, b ER-positive, and c ER-negative tissue. Edge weight indicates the statistical significance of the epithelial-stromal co-expression relationship, edge color indicates the direction of the association (positive is yellow; negative is gray), node size indicates the node’s degree, and node color indicates whether the node was involved in an epithelial-stromal self-loop (pink) or not (gray). The network layout was determined by a force-directed algorithm applied to the network and then by performing hierarchical clustering on the network adjacency matrix and projecting the clustering results onto the network. Panels d and e show the results of unsupervised hierarchical clustering performed on the ER-positive network adjacency matrix, which uncovered a network cluster highly enriched for proteins involved in the type 1 interferon response (d) and the mitotic cell cycle (e)