Fig. 3

Random Walk with Restart for multiplex/heterogeneous graphs: A and B Comparison between RWR-MH and MultiXrank for intra- and inter-layer transition probabilities as a function of switch probability, for a hypothetical source-target node pair with source node degree of 50 and source-target edge weight of 0.8 in a multiplex graph with 3 layers. C–E Empirical cumulative probabilities for ranks of nodes (determined by diffusion scores) in test folds following a fivefold cross-validation procedure using KEGG metabolic pathways, wherein each pathway is split into 5 folds, features in training folds are set as seeds, and the ranks of the features in the test fold are stored. Restart probability is set to 0.5. Ranks shown include ranks from all test folds from all KEGG pathways. Lower ranks represent higher diffusion scores and higher affinity to seed nodes. The networks used include different combinations of three components: a Mus musculus physical PPI network from STRING; a functional PPI network wherein an edge connects two nodes if they belong to a common Reactome pathway; and a MMI network of merged isomers from STITCH. The multiplex-heterogeneous graph comprises all three components, with the two PPI networks as layers in a multiplex. The monoplex-heterogeneous graph comprises the functional PPI network and the MMI network. The multiplex-homogeneous graph comprises the two PPI networks as layers