Fig. 4

The AUROCs of SRGS and each compared GRN method tested on experimental single-cell gene expression datasets, including mHSC-E, mHSC-GM, mHSC-L, mDC, hHEP, and hESC. The referenced ground-truths include cell-type-specific ChIP-seq data, nonspecific ChIP-seq data, and STRING functional interaction network. The top 500 or 1000 variable genes were used for GRN inference. A The number of overlapping TFs between the top genes in each single-cell dataset and the regulators in each referenced ground-truth, and the number of overlapping genes between the top genes in each single-cell dataset and the targets in each referenced ground-truth. The AUROCs of all methods tested on all datasets and ground-truths, when (B) top 500 genes and (C) top 1000 genes were used for GRN inference. D The mean and standard deviation of AUROC of each GRN method across all single-cell datasets, referenced ground-truths, and different numbers of top genes. The evaluated methods are SRGS, GENIE3, SINCERITIES, ARACNE, CLR, MRNET, PIDC, PPCOR, LEAP, and bnlearn