Fig. 1. Data collation (top) and modelling (bottom) workflows. The modelling workflow was repeated five times, varying which fifth of the data was held out as the external validation set such all data was included in the external set once. This five-fold cross validation was repeated 20 times with 20 different 5-way cross-validation splits. The entire procedure was repeated for each combination of descriptors in order to establish whether combining different input domains improve toxicity predictions on the dataset used in this study.
