Methods: We introduce Probabilistic Boolean Network (PBN) models topologically matched to cellular signaling networks for interpreting transcriptional responses to therapy. PBN models have the advantage of abstracting the fundamental switch-like behavior of signaling proteins, while retaining a stochastic process of activation that can successfully model a true biological system. In addition, the activity of nodes in a PBN can be elucidated by Markov chain exploration driven by determination of the probability of the changes in activity of transcription factors. The probability of the activity of a transcription factor can be deduced by statistical analysis of transcriptional data, such as provided by the use of microarray measurements of tumor or biopsy samples following application of a therapeutic. While it would be simpler to make direct phosphoprotein measurements, these generally cannot be made on a large scale for the many potential protein changes.
Results: We demonstrate our approach with two examples. First, we demonstrate how a PBN incorporating stochastic transitions governed by a beta distribution derived from signaling states in connected proteins and probabilities of transcription factor activities estimated from microarray data can successfully recover the relative level of signaling in proteins in the MAPK pathways in S. cerevisiae. Second, we show how to contruct PBNs for c-KIT signaling in gastrointestinal stromal tumors, and explore how estimates of transcription factor activity pre- and post-treatment with the c-KIT inhibitor Gleevec can be used to estimate activity of individual components.
Conclusions: PBNs offer a potential method to determine activity of specific signaling proteins following administration of targeted therapeutics designed to inhibit cellular signaling.