Figure 1.

Experimental design of the study. First we performed a prevalidation analysis of the genomic predictors for 16 anticancer drugs using cross-validation in the training set cancer genome project (CGP). We then built genomic predictors using the full training set and evaluate their performance in the cancer cell line encyclopedia (CCLE), a fully independent validation dataset. We split the validation dataset into two parts containing the cell lines that are present in the training set (COMMON) or cell lines that are unique to the CCLE dataset (NEW).