The description of cellular processes and the quantitative analysis of their perturbations is a crucial step towards understanding disease. It is through the combination of prior knowledge captured in biological network models with high-throughput experimental data that one can explore and understand how cellular processes are impacted by external stressors such as exposure to a broad variety of substances. We are using network models constructed using Selventa® knowledge assemblies to describe non-kinetic causal relationships between biological processes. In such network models, some nodes—termed “hypotheses”—are associated with a set of genes which correspond to the direct downstream targets of the process described by the node. The agreement between the behavior contained in the model and the behavior observed at the gene expression level in a particular experiment allows us to quantify the validity of the corresponding “hypothesis”. Not all biological entities in the model can be linked to experimental evidence and, when specific experimental data are gathered, the model will not be equally activated due to experimental setup specificities. This motivated the development of a methodology for soft integration of experimental data with prior knowledge about biological network topology, to identify and quantify the perturbed regions of the network. To this end, the causal network is considered a directed graph on which an irreducible simple random walk is defined. Asymptotic properties of the random walk serve to define the topological importance of each node. The topological effect of the perturbation is then quantified by reinforcing the random walk according to the collected experimental evidence about the perturbation of hypotheses in the model and comparing to the uniform case. The proposed method is applied to the NF-κB signaling network model for normal human bronchial epithelial cells exposed to varying concentrations of TNFα for various time periods. The perturbed regions are identified and correctly reveal a time- and dose-dependent response. Thus, the structure of the overall systems response hidden in the noisy behavior of thousands of downstream-controlled genes is elegantly captured by our approach. The method provides a useful way to describe global effects of external perturbations on a biological network by combining the knowledge contained in a causal model and the systems response measured by gene expression technology.