Traditionally, models in biology are used to generate predictions of unobserved responses based on assignment of parameter values. However, it is often infeasible to measure values of various biological parameters in vivo. Moreover, the mechanistic dynamic ODE models of cell signaling pathways are vastly underdetermined given limited experimental measurements. This problem then calls for the use of probabilistic approaches. We implemented framework to infer probabilistic distributions of parameter values in large-scale ODE models of cell signaling pathways and demonstrated it on example of several biological questions.
We study signaling pathways involved in cancer, and apply the probabilistic framework to fit the mechanistic model to experimental data and infer parameter distributions consistent with the data. We can then generate probabilistic predictions of unobserved behaviors of the system – assigning probability to each outcome across all range of possible values, not only a single point estimate. We also develop statistical methods to explore the relationships between model’s parameters and system responses. Further, we present how the framework can be used to dissect molecular differences between cancer and normal cell lines based on their phospho-proteomic signaling profiles.