Skip to main content
Fig. 2 | BioData Mining

Fig. 2

From: Personalized single-cell networks: a framework to predict the response of any gene to any drug for any patient

Fig. 2

Our goal is to combine a generic gene-based bipartite graph with the auxiliary knowledge of a fully-connected personalized graph. RWR will impute the missing links (and update the existing links) by “walking through” the auxiliary information. The left panel shows a (sparsely-connected) gene-drug graph combined with a (fully-connected) gene-gene graph, where di represents the drugs and gi represents the genes. The example gene-drug network has missing links. The right panel shows the output of RWR: a complete network of newly predicted gene-drug interactions. Here, the missing link between any drug di and any gene gi is replaced with a new link. The method works based on the principle of “guilt-by-association”: the value of the new digi links will be large if gi is strongly connected to genes that are also connected to di. When the gene-drug graph is fully-connected, RWR will instead “update” the importance of each connection. Note that, with respect to this figure, the gene-annotation bipartite graph is conceptually equivalent to the gene-drug bipartite graph, except that many edges will have zero weight

Back to article page