Efficient Boolean Modeling of Gene Regulatory Networks via Random Forest Based Feature Selection and Best-Fit Extension

Gene regulatory networks play a critical role in cellular behavior and decision making. Mathematical modeling of gene regulatory networks can help unravel the complexity of gene regulation and provide deep insights into key biological processes at the cellular level. In this paper, we focus on building Boolean models for gene regulatory networks from time series gene expression data. Since the two classic methods, REVEAL and Best-Fit Extension, are both computationally expensive and cannot scale well for large networks, we propose a novel hybrid approach combining the feature selection technique based on random forest and the Best-Fit Extension algorithm. The feature selection step can effectively rule out most of the incorrect candidate regulators, and thereby can significantly decrease the workload of the subsequent Best-Fit Extension fitting procedure. The efficiency and performance of the proposed two-stage framework are analyzed theoretically and validated comprehensively with synthetic datasets generated by the core regulatory network active in myeloid differentiation.