Figure 1

Method overview. (A) Two different cells (ovals): a healthy cell h in a neutral environment (left) and a damaged cell d treated with neocarzinostatin (right). Inside each oval: a pathway topology, with regulators RelA, ATM and p53, and a set of remaining genes g 1-g 4. Solid edges: signaling relations (e.g. in d, ATM signals down to RelA). Dashed edges: transcriptional regulation. Δ RelA - an experiment, where RelA is perturbed. Gene colors: effect of perturbation (up-regulation in red, down-regulation in green and no change in white). (B) The JODA algorithm. Input: (i) perturbation data, (ii) known TF-targets, and (iii) known pathway models encoded in matrices with an entry 1 when a perturbation experiment (rows) affects the regulator (columns; otherwise the entries are 0). Experiments affecting RelA are marked in blue. The input is processed for the healthy (left) and the damaged cells (right) separately in three steps, until merged in deregulation scores. Examples on the right illustrate the steps for RelA. First step: 'bgmm' [32] is applied to identify probabilities of differential expression of the genes under perturbation of each regulator v (denoted and for the two cell populations h and d). Each probability is multiplied by -1 when the effect of perturbation was down-regulation, or by +1 when the effect was up-regulation. Second step: we compute regulation scores and , which quantify the effect of each regulator v on the genes in a given cell population. Third step: we subtract regulation scores in the healthy cells from regulation scores in the damaged cells to obtain deregulation scores D _v , quantifying how strongly each regulator v deregulates the genes.