As a research on molecular design, corresponding to “What to Make”, we focus on three dimensional quantitative structure-activity relationships (QSAR) by chemoinformatics approaches. For instance, in order to specify substructures or region that gives important effect on activity of medicine, we have proposed a new technique for region selection (variable selection) based on the genetic algorithm. This technique is applied to designing of high activity compounds with pharmaceutical companies. Moreover, the multi-way PLS method, which can make a robust model when data structure of the explanatory variable is three-dimensions or more, is applied to construction of QSAR models. Furthermore, we have succeeded in proposing an epoch-making technique for identifying significant area on molecular surface for high activity. The method uses electrostatic potential and hydrophobic parameter on molecular surface of a series of compound. We have used Kohonen self-organizing neural network and multi-way PLS method to cope with these data on molecular surface that has different shape. With these prediction models, we are developing the method to automatically generate three-dimensional structures of medicines having higher activity.
Besides, we are developing the total molecular design system as the integration system for these analyses. This figure shows a screen shot of this system. In addition, we are researching not only QSAR analysis and drug design but also catalyst design, polymer design and so on.