Fig. 1.

Ensemble effects on class-imbalanced datasets. a Improved average AUC value produced by neural network bagging (NN-bagging) and neural network-based representation ensemble (NN-representation ensemble) over three fingerprints. b Pearson’s correlation (r=0.69, p-value=1.1x 10⁻³) between the improved AUC values from NN-bagging and the class imbalance ratio. The class imbalance ratio was calculated from the number of active and inactive chemicals, as shown in Table 1