Fig. 4
From: Uncovering cancer vulnerabilities by machine learning prediction of synthetic lethality
Vulnerabilities of DDR related genes. a Network of predicted SLs between DDR genes and the genome based on the raw permutation importance score. Each node represents a gene and each edge a relationship; the arrow starts from the mutated (green) or dysregulated gene (blue) and arrives to the dependent gene. The width is proportional to the absolute value of the Pearson correlation coefficient. The color of the node shows the median of the dependency score of the gene in a grey scale. Different arrow shapes show three levels of confidence based on the scaled importance score. b Scaled raw permutation importance score distributions of selected gene pairs divided into paralogs or not in the two cohorts (expression and mutation). c Bar plots showing examples of high-confidence predicted SL gene pairs using expression features. The ranked bars show the dependency scores (mean centered) of one gene across the cancer cell lines and the color gradient shows the expression level of the second gene. d Scatterplots showing the gene expression levels (based on RNA-Seq) of BRIP1 and ALDH2 in matched tumor (TCGA) and normal (GTEX) breast, lung and brain tissues. e Top altered pathways in the enrichment analysis of differentially expressed genes in TCGA cc samples expressing high or low ALDH2. f Heatmap showing expression levels (mean centered) of the main Fanconi anemia genes in TCGA breast cancer samples expressing high or low ALDH2. g Scatterplots indicating the correlation between the expression levels of ALDH2 and BRIP1 gene effect together with the promoter methylation levels of ALDH2 in breast cancer cell lines