Fig. 7

Heatmap of the non-normalized ensemble spread of QSAR model ensemble predictions for reduction potentials of generated quinone structures based on the template quinone 1,2-naphthoquinone with a maximum of two hydroxyl and methyl groups at varying positions. Ensemble predictions are obtained through 10-fold cross-validation models trained on a data set of 69,599 quinones. Fields marked with ‘*’ are quinones that are present in the training data set. White fields are impossible quinone structures. The molecular structures associated with the four largest ensemble spread values are shown in the bottom left