Figure 2

The predictive performance of BICEPP (AUC) by drug characteristics. The predictive power of BICEPP was evaluated by using stratified cross-validation experiments performed on each of the 484 drug characteristics listed in Table 1. In this figure, each data point denotes the best AUC (out of the 4 machine learning algorithms) evaluated on the dataset of drug characteristic that contains more than 10 positive examples (O), between 2--9 positive examples (*), and less than 2 positive examples (X) respectively. The dotted lines indicate AUCs of 0.8 and 0.9 respectively. Abbreviations: TI: therapeutic index.