Simplify regulator-target gene network via clustering.
Several regulators (specifically TFs) might have extremely overlapping target genes. In order to identify clusters of highly similar regulators (mainly TFs) we implemented a network simplification algorithm in biRte: We construct the biadjacency matrix of the complete bipartite regulator target-gene graph and then calculate a single linkage clustering of regulators based on the Tanimoto-Jaccard similarity of their target genes. The dendrogram is cut at a defined height (default: 0.1) to idenfity resulting groups. The algorithm is meant to simplify the inference of active regulators, because the resulting regulator clusters have more dissimilar target gene profiles.
original regulator-target gene network
cut dendrogram at height 1 - cutoff (i.e. similarity cutoff)
clustered / simplified network
1 2 3
# artificial data data(humanNetworkSimul) affinities2 = simplify(affinities2)