Building model prototypes from time-course data

A primary challenge in building predictive models from temporal data is selecting the appropriate model topology and the regulatory functions that describe the data. Software packages are available for equation learning of continuous models, but not for discrete models. In this paper we introduce a method for building model prototypes. These model prototypes consist of a wiring diagram and a set of discrete functions that can explain the time course data. The method takes as input a collection of time course data or discretized measurements over time. After network inference, we use our toolbox to simulate the prototype model as a stochastic Boolean model. Our method provides a model that can qualitatively reproduce the patterns of the original data and can further be used for model analysis, making predictions, and designing interventions. We applied our method to a time-course, gene-expression data that were collected during salamander tail regeneration under control and intervention conditions. The inferred model captures important regulations that were previously validated in the research literature and gives novel interactions for future testing. The toolbox for inference and simulations is freely available at github.com/alanavc/prototype-model .