Fig. 2: The performance of iGenSig models in predicting the drug responses of GDSC cell lines.

a The performance of the iGenSig models for GDSC profiled drugs was assessed by their average AUROC. About 364 drugs that show a negatively skewed drug response distribution in cancer cell lines and have at least 20 sensitive cell lines are included in the analyses. The drugs with top-performing models (AUROC >0.85) are shown in bar chart on the right. The average AUROC for each drug was calculated based on five train/test sets. b Correlating the performance of the iGenSig models for 364 drugs with their average number of significant genomic features. The drug models assessed on the six clinical trial datasets are highlighted in red. c the performance of the iGenSig model for Lapatinib in predicting the response of GDSC cell lines based on a representative training and testing set. Left, sensitive, and resistant GenSig scores for GDSC cell lines. Middle, the correlation of the iGenSig scores with AUC measurements for Lapatinib. Right, the receiver operating characteristic (ROC) curve for predicting sensitive responses to Lapatinib. As the golden standard for the ROC curve, the cell line subjects in the test set are divided into sensitive and non-sensitive groups based on the AUC measurements for Lapatinib using the cutoff determined by the waterfall method (see Methods). d Ward D2 Hierarchical Clustering for GDSC cancer cell lines and targeted kinase drugs of five RTK signalings based on iGenSig scores. The drugs targeting different kinases or different kinase families form distinctive clusters.