Fig. 3: Predictive values of iGenSig models developed from the GDSC pharmacogenomic dataset on the sensitivity of CCLE profiled cell lines to the shared drugs.
From: An integral genomic signature approach for tailored cancer therapy using genome-wide sequencing data
a The performance of GDSC iGenSig models in predicting the responses of GDSC testing cell lines and CCLE cell lines to 14 drugs shared between the two datasets. 80% of GDSC cell lines are used for building the iGenSig models and 20% of GDSC cell lines are used for testing. 100% of CCLE cell lines are used for cross-dataset validations. If the same drug is profiled by both GDSC batch 1 and 2, the drug sensitivity data from batch 1 are used in the analysis. The error bars show standard deviations. b The predictive values of the iGenSig models developed from GDSC data on the CCLE cell lines treated with Irinotecan, Nilotinib, Lapatinib, or Erlotinib. The upper panel shows the correlation between the iGenSig scores and the Act Areas of the respective drugs for CCLE cell lines. The horizontal dashed lines show the cut-offs for sensitive (red) and resistant (blue) calls. The vertical dashed line shows the optimal cut-off for iGenSig scores determined based on AUROC. The lower panel shows the ROC curves of iGenSig scores in determining the sensitive cell lines vs non-sensitive cell lines. c GDSC and CCLE cell lines show consistent integral genomic signature that correlates with Erlotinib responses. The significant genomic features (n = 8540) based on K-S tests are shown in the figure. The GDSC and CCLE cell lines are first sorted by their sensitive iGenSig scores; the cell lines with sensitive iGenSig scores less than the median are then sorted by the resistant iGenSig cores. The cell lines that have been tested for Erlotinib chemical perturbations are shown in the figure, and the sensitive and resistant cell line subjects are indicated as yellow and blue bars.
