Development of new 3D-QSAR method by Kohonen network and 3wayPLS analysis

The quantitative description of chemical structures and their activities is called quantitative structure-activity relationships (QSAR). Recently, Comparative Molecular Field Analysis (CoMFA) is more often used. In the CoMFA modeling, steric and electrostatic interaction energies between molecules and probe atoms are used for the modeling. And the relationship between these interaction energies and biological activities are modeled using partial least squares (PLS) method. Therefore, three-dimensional structural features of the molecules can be directly reflected to the CoMFA model. However, conventional CoMFA have some problems; for example, because of the CoMFA model has a large number of descriptor variables, the model would become unstable. If advantages are mentioned, it is possible to model efficiently. In this study, we developed a new 3D-QSAR method. We have analyzed relationship between each Molecular electrostatic potential (MEP) on the van der Waals surface and their activities on 25 Molecules that are various substituted salicylamides, having 2-pyrrolidinylmethyl side chains. The compounds belong to the class of substituted benzamides and display potent binding to the dopamine D-2 receptor. MEP is important information for studies on pharmacologinal activity. The three-dimensional of the MEP is largely responsible for binding of ligand molecule at the active site of receptor. First each molecular three Cartesian coordinates of points on its van der Waals surface and MEP values of those points are obtained, and train Kohonen neural network with its three Cartesian coordinates. Then two-dimensional maps of MEP for 25 molecules are obtained. The characteristic of molecule surface is able to be expressed well by Kohonen network. Then they are analyzed by PLS method or 3wayPLS method. The result suggested that 3wayPLS method could show the effective regions for activity more clearly than 2wayPLS method did. Last all data was divided into training data and test data by D-optimal designs and test data were predicted. The prediction result suggested that 3wayPLS method is better than 2wayPLS method too.