To extract modules from compendia of microarrays, we largely followed the knowledge-driven approach proposed by Segal et al. [10] (Figure 1, Step 3).
All direct interactions of the prioritized TFs were used to extract modules.
We extract modules for 8,098 out of 12,621 GO BP terms represented in the STRING network.
A first examination showed that CFinder could extract modules that map to different sub-units of large complexes.
Many network-based analysis methods extract modules, which are sets of tightly connected nodes consisting of the input genes, and it is strongly expected that the genes in a module achieve a biological function in a coordinated manner.
Unfortunately, for any sensible definition of modularity, graph-based module identification is typically computationally very expensive, so in practice one resorts to heuristic algorithms to extract modules [ 49, 51, 52].
Since the goal of alignment is to extract modules that correspond to specific biological processes, we examined the use of functional similarity of proteins across networks to aid alignment.
We observed that CFinder could extract modules which match well with multiple MIPS complexes, and in some cases also functionally related to other proteins (which we found with STRING, a tool based on different type of evidences with a corresponding score), or instead with relatively poor match with MIPS.
Similarly to EEM, when given a TF binding motif gene set, BEEM predicts genes under a common cis-regulatory code as an expression module; Furthermore, the extracted module information can be used to inspect the upstream transcriptional program.
In some cases, we observed that a module may represent the residue or remaining signal of a stronger, previously extracted module.
