Binary QSAR-based library design

COMP 168

Hua Gao,, Computational Chemistry, Pfizer Inc, Eastern Point Road, Groton, CT 06340
A set of 671 cdk5 inhibitors was analyzed with binary QSAR method. A binary QSAR model with predictive accuracy of 90% on actives, 91% on inactives (overall 91%) was obtained for a training set of 492 compounds. The model was validated with a set of 51 compounds not included in the training set, and further tested with a set of 128 inactives (IC50 > 100 mM). The derived binary QSAR model was successfully used in the screening of a virtual combinatorial library of 300,000 compounds. Good results were obtained when compounds were selected from the virtual library based on the binary QSAR model for synthesis and subsequently tested against cdk5. The result indicates that binary QSAR is very useful in compound selection for HTS, and in focused or targeted combinatorial library design.