Fig. 1. Illustration of the computational framework for gene regulatory network inference, optimization, and modeling.

a Procedures for gene regulatory network (GRN) inference and optimization. The top left block shows the steps to construct TF-target databases (DB) using a literature-based TF-target DB and the TF-target relationships inferred from ATAC-seq data. The top right block shows the approach of TF inference using three distinct computational methods: VIPER, RI regression, and NetAct. The bottom block shows the steps to construct GRN candidates using the TF-target databases and TF activities. Many candidate GRNs are constructed by varying three adjustable hyperparameters, as highlighted in red color. Network optimization is then applied to identify the optimal GRN that best captures experimental gene expression states according to GRN simulations by RACIPE, while also maintaining flexibility to allow network state transitions. b Network annotation. Using the GRN-related differentially expressed genes (DEGs), we identify enriched KEGG biological pathways and the best representative pathways associated with each network TF. c Network dynamics characterization. In silico network perturbation analysis can be further performed to identify key regulators of the network driving state transitions.