Integration of identified CRISPR hits and fitness signature to human AML. (A) Overview of computational approach. For the human AML gene expression data sets, a transcriptional regulatory network was constructed with ARACNe, and the protein activity of each MR in each sample was assessed by VIPER. Patient samples were stratified by enrichment of the LSC signature. The top 50 patients with highest enrichment were compared with those with lowest enrichment using VIPER to identify potential upstream MRs with an FDR cutoff of 5%. (B) MRs predicted to be activated in human AML samples with higher LSC, grouped according to whether they were homologs of CRISPR hits in vitro, in vivo, or in both conditions. Only homologs of genes that are covered in the secondary CRISPR screen are shown. (C) Prognostic analysis of CRISPR hits. The y-axis in the bar plot shows the fraction of human homologs of CRISPR hits that are significantly associated with survival in at least 1 of the 4 AML data sets (P values <.01), according to classification of CRISPR hits. (D) Homologs of in vivo–only CRISPR hits that are also associated with AML survival. (E) In each of the 4 AML data sets, patients were stratified by expression of FERMT3, and the top 25% samples with highest expression of FERMT3 were compared with the bottom 25%. Hazard ratios (HRs) and log-rank test P values are also shown.