Tree-based learning of regulatory network topologies and dynamics with Jump3

Inference of gene regulatory networks (GRNs) from time series data is a well established field in computational systems biology. Most approaches can be broadly divided in two families: model-based and model-free methods. These two families are highly complementary: model-based methods seek to identify a formal mathematical model of the system. They thus have transparent and interpretable semantics, but rely on strong assumptions and are rather computationally intensive. On the other hand, model-free methods have typically good scalability. Since they are not based on any parametric model, they are more flexible that model-based methods, but also less interpretable.
In this chapter, we describe Jump3, a hybrid approach that bridges the gap between model-free and model-based methods. Jump3 uses a formal stochastic differential equation to model each gene expression, but reconstructs the GRN topology with a non-parametric method based on decision trees. We briefly review the theoretical and algorithmic foundations of Jump3, and then proceed to provide a step by step tutorial of the associated software usage.