Fig. 1

scMAGeCK pipeline and a comparison with clustering analysis and other methods on single-cell CRISPR screens. a An overview of the scMAGeCK pipeline. The input of scMAGeCK includes a scaled expression matrix of all genes in all single cells, together with cell identity information on the targets of each single cell. scMAGeCK includes two modules: RRA and LR. RRA infers gene regulatory relationship on certain gene expression (e.g., gene A) using the rankings of single cells and takes dropout events into consideration. LR infers the gene regulatory network on all possible gene expressions. b A comparison of scMAGeCK with clustering analysis on three different public CROP-seq datasets. The total number of target genes, genes that are enriched in certain cluster, and genes whose downregulation is considered as statistically significant (FDR < 0.25) are shown. Gene A is considered enriched in certain cluster are defined as single cells carrying gene A knockout consists of > 20% total cells in that cluster, and with adjusted p value smaller than 0.25 using chi-squared test. c The ranking of genes in reducing CD3D expression in the T cell CROP-seq dataset. d The significant GO terms (FDR < 0.05) in the permutated CROP-seq datasets as a measurement of false positives. e The significant genes (FDR < 0.05) of each method in the permutated CROP-seq datasets. For all the datasets, we randomly selected 50 expression markers and identified significant perturbations as a measurement of false positives. f The selection score distribution of scMAGeCK-LR and MIMOSCA over 145 validated enhancer-gene pairs in [13]. The number of pairs identified by each method is shown in parenthesis