Fig. 1: Overview of the CEFCON framework.

a CEFCON takes a prior gene interaction network and gene expression profiles derived from the scRNA-seq data of a given cell lineage trajectory as inputs. b A gene regulatory network (GRN) related to a specific cell lineage is first constructed through contrastive learning on a graph neural network (GNN) with attention mechanism. CEFCON uses the attention coefficients to assign weights to individual gene interactions and then selects the top-weighted interactions to construct the cell-lineage-specific GRN. c CEFCON uses network control-based methods to identify driver regulators that steer cell fate decisions along the developmental trajectory. d, CEFCON identifies the regulon-like gene modules involving the selected driver regulators.