Literature and data-driven based inference of signalling interactions using time-course data

Abstract Cellular activity and responses to stimuli are governed through an elaborated communication process called cell signalling. The modelling of signalling mechanisms has the potential to help us understand the regulatory processes determining cellular behaviour. One approach to derive models of signalling networks is from data alone. Another one is to use prior knowledge networks (PKN’s) derived from literature or experts’ knowledge to build models that are trained to data. Both approaches have limitations. Data-driven methods can infer many false-positive interactions. Literature-constrained methods, on the other hand, are limited to model only known interactions. To overcome these limitations, within a logic ordinary differential equations (ODE) formalism, we have developed Dynamic-Feeder. The framework identifies and incorporates new possible links to the network and then it evaluates their effects based on how the models predict the data. Dynamic-Feeder combines data-driven inference methods with general literature-based knowledge of proteins interaction networks (PIN’s). We illustrate our method with a published case study using phosphoproteomic data upon perturbation of breast cancer cell lines.