Figure 7

Confidence intervals calculated from the ensemble standard deviation of the models present in the model ensembles. The percentage of datapoints which predicted bioactivities lie within confidence intervals calculated with increasingly larger β values (Equation 4), is shown for: (i) the cross validated predictions calculated during model training (Training in the Figure), and (ii) for the predictions on the test set (Test in the Figure) calculated with the most predictive model ensemble, namely “Stacking SVM Radial Ensemble”. The percentage of true values lying within the confidence interval derived for a given β value increases with the number of datapoints available during model training. Overall, the confidence intervals derived from the ensemble standard deviation provide an estimation of the reliability of individual predictions, as in practice, this plot can be used to determine the β value corresponding to a given confidence level.