Figure 3
From: Reliable anti-cancer drug sensitivity prediction and prioritization
Classification test set performance GDSC2. The upper row of this figure depicts the classification performance of SAURON-RF across the different drugs from GDSC2. The notation on the x-axis of the first plot consists of a tuple containing the true class as first element and the predicted class as second element. For all predictions where the true class is sensitive (i.e., TP or FN), percents are calculated by dividing by the number of all sensitive cell lines (TP + FN). Likewise, for all predictions where the true class is resistant (i.e., TN or FP), percents are calculated by dividing by the number of all resistant cell lines (TN + FP). Thus, the x-axis labels correspond to the well-known confusion matrix metrics called sensitivity = \(\frac{\text {TP}}{\text {TP + FN}}\), miss-rate = \(\frac{\text {FN}}{\text {TP + FN}}\), specificity = \(\frac{\text {TN}}{\text {TN + FP}}\), and fall-out = \(\frac{\text {FP}}{\text {TN + FP}}\), respectively. The middle row shows the effects of CP on the performance in terms of true positive/negative predictions. Again, tuples of the true and the predicted class sets are shown on the x-axis and percents were obtained as described above. In Supplement 1 Section 7, we provide all formulas. In the lower row of this figure, the CP efficiency is presented.
