Table 2 Synthetic data sets to benchmark interpretation of QSAR models
Dataset | Property type | End-point | Train/test set size | Expected atom contribution |
|---|---|---|---|---|
N | Regression | Sum(N) | 6995/2999 | Nitrogen atoms: 1; others: 0 |
N − O | Regression | Sum(N) − sum(O) | 6893/2969 | Nitrogen atoms: 1; Oxygen atoms: − 1; others: 0 |
N + O | Regression | (Sum(N) + sum(O))/2,where sum(N) = sum(O) | 7000/3000 | Nitrogen and Oxygen atoms: 0.5; others: 0 |
Amide_reg | Regression | Sum(NC=O) | 7000/3001 | Any atom of amide groups: 1; others: 0 |
Amide_class | Classification | Active: if sum(NC=O) > 0; inactive: if sum(NC=O) = 0 | 6998/3000 | Any atom of amide groups: 1; others: 0 |
Pharmacophore | Classification | Active: at least one conformer with exactly one pharmacophore match (same two atoms in all conformers); inactive: no pharmacophore matches for all conformers; pharmacophore match: HBD and HBA 9–10 Å apart | 7000/3000 | Atoms which are HBA or HBD of the pharmacophore: 1; others: 0 |