Table 3.

Evaluation methods for in silico models

Modeling type	Description of the method
For all modeling work:	▪ Standardization and curation of the investigated dataset to ensure consistency. This should include a clearly-stated method (including inclusion and exclusion criteria) for curation of the data and a review of the rules applied to chemical structures in order to ensure standardization
QSAR models:	▪ Use of sufficiently diverse training set covering the EDC compound domain of interest
	▪ Use of sufficiently diverse external test set covering the EDC compound domain of interest should be used
	▪ Assembly of internal and external validation, i.e. several internal and external validation sets, and models created in a double loop fashion, followed by consensus predictions
	▪ Sufficient statistical quality achieved
	▪ Consistent applicability domain established, e.g. using a conformal prediction framework
For ligand based pharmacophore models:	▪ Use of sufficiently diverse training set covering the EDC compound mechanism/domain of interest
	▪ All training set compounds should, approximately, fit the derived model equally well unless there are demonstrable differences in the binding affinity
	▪ Use of sufficiently diverse external test set that covers the EDC compound domain of interest to demonstrate generalizability
Protein structure based models:	▪ Several protein structures should be used to account for flexibility of the protein covering relevant conformations
	▪ Use of sufficiently diverse training set covering the EDC compound domain of interest
	▪ Consensus docking and scoring to ensure robustness and stability of results
	▪ Use of sufficiently diverse external test set covering the EDC compound domain of interest