should be included into the model (seeNote 3); weighing the
contribution of such descriptors to the modeled activity; validating
the model internally and externally and; checking the applicability
domain of the model whenever a prediction is made [34]. Molecular
diversity of the training samples is critical for VS applications of
supervised machine learning: the molecular diversity of the calibra-
tion examples is directly correlated with a wide applicability domain
of the resulting model.
Fig. 2Snapshots from a molecular dynamics simulation of the interaction
between anticonvulsant sulfamides and carbonic anhydrase. Note the
significant conformational changes induced by the ligand binding event
Computer-Aided Drug Design 7