Figure 4

Estimating the uncertainty profile for the Ames data set. The validation set results shown in panels A and B are for a model (Ames-1) having two hidden neurons and 26 descriptors as input, whereas the results shown in panels C and D are for a model (Ames-2) having four hidden neurons and taking 24 descriptors as input. Different random number seeds were used to split the respective training pools into training and test sets and to initialize the individual ANN weights. The voting threshold for both (indicated by vertical black dotted lines) was 16.5. (A and C) Distribution of predictions (blue) and errors (red) for the external validation set. Dashed lines represent the fitted beta binomial distributions obtained from the corresponding training pool results. (B and D) Observed (red line) error rate profile for the validation set and predicted uncertainty profile (dashed black line) based on the training pool.