Fig. 1
Schema describing the pipeline for building networks and predicting regulatory nodes. (1) Using a list of differentially expressed genes, construct the set of gene names and the corresponding set of observations (a sign is attributed for each gene: + when fold change >2 or − when fold change <0.5 and adjusted p-value < 10−5); (2) Extract the upstream/downstream signaling pathways for the set of genes from the signed interaction graph using Pathrider, a tool developed in our team to this purpose. Given a list of excluded genes (such as invariant genes), Pathrider filters these genes to reduce the graph size; (3) Check the sign consistency of our datasets to produce signed predictions for unmeasured biomolecules using iggy tool; (4) Validate the predictions made by iggy by computing sub-predictions (prediction 1, 2...n) using a sub-set of observations (by default, it starts sampling from 10% to 95% of observations with a step of 5% and a number of execution equal to 100), then compare it firstly with the differentially expressed genes, and in a second time with the predictions obtained with all the set of observations; and (5) Plot the precision scores for each sub-sets of the observations, and the stability of the prediction compared to the predictions of the entire set of observations