Figure 1. Measuring the performance of the most significant single-gene marker for the drug AZD7762.

Using all the cell lines originally tested with this drug (the training set), MYCN-mutated cell lines were found to have increased sensitivity to AZD7762 (P=0.002)⁶. The median logIC50 in μM units of all cell lines in this training set (left) defines the sensitivity threshold (horizontal red line) for both training and test sets. Thus, this single-gene classifier predicts that cell lines harbouring the genetic aberration (mutated MYCN with boxplot named MYCN for short), tend to be more sensitive to AZD7762 (logIC50 below threshold) than cell lines that are wild-type (WT) with respect to the MYCN gene (boxplot marked as WT for short). This classifier achieves a high precision (PR=0.75) on the training set because only 25% of the cell lines that are predicted to be sensitive are actually resistant (false positives; FP). However, it obtains a very low recall (RC=0.05) as the vast majority of sensitive cell lines are WT with respect to MYCN and hence are incorrectly predicted to be resistant (i.e. false negatives; FN). A second scatter plot with two boxplots (right) shows the classification performance of the AZD7762-MYCN marker on the test set. It is similar to the performance obtained on the training set, which evidences the robustness of this particular marker.