Figure 8

Averaged ROC curves for Set A aryl-amine PLS models. The plot on the left is for the model built with a full descriptor set and the plot on the right is for a limited descriptor set. The left plot shows the averaged (identical false positive values) ROC curves for the test (dark line) and training (lighter line) sets of 100 PLS models built on a random 70% sample of the Set A aryl-amine data (MW < 250 g/mol) and the performance of a PLS model built on all of the data as a dashed line with all non-zero-variance descriptors. The right plot uses only 9 descriptors including nitrenium formation energy.