Skip to main content
. 2020 Feb;30(2):227–238. doi: 10.1101/gr.250316.119

Figure 1.

Figure 1.

Workflow of our MOCHI algorithm and output examples of HIMs. The network has both gene–gene spatial proximity and TF–gene regulation relationships. (A) A four-node motif M represents the smallest HIM. Here a directed interaction represents a TF–gene regulation relationship, and an undirected interaction represents that the two genes are spatially more proximal to each other than expected. (B) Given a heterogeneous network G, we find HIMs by minimizing the motif conductance (see Equation 2). (C) We compute the subgraph adjacency matrix WM, with [WM]ij being the number of occurrences of M that have both nodes i and j. (D) The weighted network GM is defined from adjacency matrix WM. (E) Spectral clustering will find clusters in GM. We recursively apply the method to find multiple HIMs and overlapping HIMs. (F,G) Two HIMs as examples in GM12878. (H) Example of two overlapping HIMs in GM12878 sharing seven TFs (the group with pink nodes in the middle). TFs in orange and pink nodes form one HIM with their target genes (bottom left). TFs in pink and blue nodes form another HIM with their target genes (bottom right). Note that the directed interactions from TFs to their target genes are bundled.