Comparison of binary and 2D QSAR analyses using inhibitors of human carbonic anhydrase II as a test case

Hua Gao
Computational Chemistry and Informatics, MDS Panlabs, 11804 North Creek Pkwy S., Bothell, WA 98011, U.S.A.

Jürgen Bajorath
Computational Chemistry and Informatics, MDS Panlabs, 11804 North Creek Pkwy S., Bothell, WA 98011, U.S.A.
and Department of Biological Structure, University of Washington, Seattle, WA 98195, U.S.A.

Abstract
Binary and conventional 2D QSAR have been derived for a set of carbonic anhydrase II (CA II) inhibitors. An overall predictive accuracy of 94% was obtained by binary QSAR and of 84% by 2D QSAR model. For both models, preferred molecular descriptor sets were identified, which were overlapping but not identical. Both binary and 2D QSAR captured important molecular features of CA II inhibitors, notably the presence of a sulfonamido group, which is critical for binding, but also hydrophobicity. Promising results were obtained when the derived QSAR models were used to test a set of CA II inhibitors not included in the training set. In binary QSAR, previously unobserved boundary effects were detected both in the analysis of known inhibitors and when screening a large combinatorial library for putative inhibitors. The complementary use of binary and conventional 2D QSAR is thought to increase the accuracy of the lead discovery process by QSAR techniques.

Keywords
binary QSAR, carbonic anhydrase, inhibitors, PLS analysis