Fig. 5

Biased random walk: impact of biased random walk with restart (B-RWR) on diffusion scores and on AMEND modules. Seed values are −log₁₀(p-value), with p-values coming from differential expression analysis of 5 human gene expression datasets. For panels B–D, results are averaged across datasets. In the context of AMEND, nodes of interest (NOI), chosen to be nodes that were removed at varying points in the iterative AMEND algorithm, were assigned large BRW attribute values. A 10% of nodes in the PPI network, chosen randomly, were assigned BRW attribute values from an exponential distribution with varying rate parameter, while all other nodes were set to 1. Y-axis shows the change in empirical cumulative probabilities for diffusion scores of these selected nodes between RWR and B-RWR, for 5 gene expression datasets. Restart parameter is set to 0.5. B B-RWR in the AMEND algorithm. All nodes were assigned BRW attribute values that decrease as a function of distance to NOI in the PPI network. Upper plot shows the change in average iteration-of-removal of NOI from AMEND_RWR to AMEND_B-RWR. Lower plot shows the change in average distance (in full PPI network) between module nodes and NOI from AMEND_RWR to AMEND_B-RWR. C–D Similar to panel B. The BRW attribute values assigned to NOI are drawn from an exponential distribution with varying rate parameters, with all other nodes set to 1