Fig. 1: Epiregulon constructs GRNs to infer regulator activity at the single-cell level.

a Epiregulon can infer regulator activity for lineage development, drug perturbations, motif-lacking co-regulators or regulators harboring neomorphic mutations. Correlation and mutual information weight estimation methods are appropriate for the first scenario, whereas co-occurrence is applicable to all cases. b If users provide scRNA-seq and scATAC-seq, Epiregulon can construct GRNs either from ChIP-seq data or motif annotations. Pan-cell-type, tissue-specific and sample-specific ChIP-seq data compiled from ChIP-Atlas and ENCODE are available through the scMultiome package. Epiregulon outputs regulator activity at the single cell level, a pruned and weighted gene regulatory network and differential activity analysis to identify potential drivers of cell states. c For benchmarking, we downloaded the paired scATAC-seq and scRNA-seq PBMC data from 10x Genomics. We identified cell types using SingleR and known marker genes. Shown is the UMAP representation of the various cell types present in the data. d Gene expression of known lineage markers. e Activities of the same lineage markers were calculated using Epiregulon (correlation weight estimation method). f Area under the receiver operating characteristic curve (AUROC) is calculated based on whether TF expression or TF activity can distinguish cells of the matching lineage vs. the remaining cells based on a total of 20 factors. g Gene expression changes after depletion of 7 individual factors (ELK1, GATA3, JUN, NFATC3, NFKB1, STAT3 and MAF) were obtained from the knockTF database and genes with absolute logFC > 0.5 and corrected p-value < 0.05 (two-sided moderated t-test, limma) were considered ground truth target genes. GRNs obtained from the shown packages were evaluated for precision and recall of target genes. h Run time and memory use of the GRN construction from the PBMC data were evaluated with 64GB and 20 cores on the high-performance computing (HPC) cluster. In the case of GRaNIE, the memory allocation needs to be increased to 128 GB and for FigR, the memory allocation was increased to 256 GB. Each package was run 5 times. Source data are provided as a Source Data file. Created in BioRender. Yao, X. (2025) https://BioRender.com/x50fdft.