Optimal Inference of Asynchronous Boolean Network Models

Associations between phenotype and genomic and epigenomic markers are often derived by correlation. Systems Biology aims to make more robust connections and uncover broader insights by modeling the cellular mechanisms that produce a phenotype. The question of choosing the modeling methodology is of central importance. A model that does not capture biological reality closely enough will not explain the system’s behavior. At the same time, highly detailed models suffer from computational limitations and are likely to overfit the data. Boolean networks strike a balance between complexity and descriptiveness and thus have received increasing interest. We previously described an algorithm for fitting Boolean networks to high-throughout experimental data that finds the optimal network with respect to the information in a given dataset. In this work, we describe a simple extension that enables the modeling of asynchronous dynamics, i.e. different reaction times for different network nodes. Our approach greatly simplifies the construction of Boolean network models for time-series datasets, where asynchronicty often occurs. We demonstrate our methodology by integrating real data from transcriptomics experiments, and provide an implementation that can be used by the community for network reconstruction using any high-throughout dataset. Our approach significantly expands the applicability of the Boolean network model to experimental data.