Fig. 3From: Predicting drug response of tumors from integrated genomic profiles by deep neural networksAssociations of gene mutations to predicted drug response in TCGA – per-cancer study. (a) Predicted IC50 of TCGA tumors with known drug targets in a cancer type. Significance of ΔIC50 between tumors with and without a gene mutation was assessed by the two-tailed t-test. (b) Significant mutation–drug pairs in each cancer with Bonferroni adjusted t-test P < 1.0 × 10−5. Nodes labeled with names are those with extreme significance (adjust P < 1.0 × 10−60) and magnitude of ΔIC50 (|ΔIC50| ≥ 0.5). Top 10 cancer types with the largest sample sizes are denoted by node color and shape. (c) Box plots of three mutation–drug examples in BRCA and LGGBack to article page