Application of the Novel Molecular Alignment Method Using the Hopfield Neural Network to 3D-QSAR
Masamoto Arakawa, Kiyoshi Hasegawa, Kimito Funatsu · Journal of Chemical Information and Computer Sciences · 2003
Recently, we investigated and proposed the novel molecular alignment method with the Hopfield Neural Network (HNN). Molecules are represented by four kinds of chemical properties (hydrophobic group, hydrogen-bonding acceptor, hydrogen-bonding donor, and hydrogen-bonding donor/acceptor), and then those properties between two molecules correspond to each other using HNN. The 12 pairs of enzyme-inhibitors were used for validation, and our method could successfully reproduce the real molecular alignments obtained from X-ray crystallography. In this paper, we apply the molecular alignment method to three-dimensional quantitative structure-activity relationship (3D-QSAR) analysis. The two data sets (human epidermal growth factor receptor-2 inhibitors and cyclooxygenase-2 inhibitors) were investigated to validate our method. As a result, the robust and predictive 3D-QSAR models were successfully obtained in both data sets.