| Abstract
| - We have developed a drug discovery strategy that employs variable selection quantitativestructure-activity relationship (QSAR) models for chemical database mining. The approachstarts with the development of rigorously validated QSAR models obtained with the variableselection k nearest neighbor (kNN) method (or, in principle, with any other robust model-building technique). Model validation is based on several statistical criteria, including therandomization of the target property (Y-randomization), independent assessment of the trainingset model's predictive power using external test sets, and the establishment of the model'sapplicability domain. All successful models are employed in database mining concurrently; ineach case, only variables selected as a result of model building (termed descriptor pharmacophore) are used in chemical similarity searches comparing active compounds of the trainingset (queries) with those in chemical databases. Specific biological activity (characteristic ofthe training set compounds) of external database entries found to be within a predefinedsimilarity threshold of the training set molecules is predicted on the basis of the validatedQSAR models using the applicability domain criteria. Compounds judged to have high predictedactivities by all or the majority of all models are considered as consensus hits. We report onthe application of this computational strategy for the first time for the discovery of anticonvulsant agents in the Maybridge and National Cancer Institute (NCI) databases containingca. 250 000 compounds combined. Forty-eight anticonvulsant agents of the functionalized aminoacid (FAA) series were used to build kNN variable selection QSAR models. The 10 best modelswere applied to mining chemical databases, and 22 compounds were selected as consensushits. Nine compounds were synthesized and tested at the NIH Epilepsy Branch, Rockville,MD using the same biological test that was employed to assess the anticonvulsant activity ofthe training set compounds; of these nine, four were exact database hits and five were derivedfrom the hits by minor chemical modifications. Seven of these nine compounds were confirmedto be active, indicating an exceptionally high hit rate. The approach described in this reportcan be used as a general rational drug discovery tool.
|
