Anwendung “künstlicher Intelligenz”in der Chemie;Computersimulationen von Struktur‐Eigenschafts‐Beziehungen

Abstract

The proceedings of chemical structure descriptors for structure‐activity studies is proposed. These descriptors are based on the set of properties distributed in a topological space (molecular graph). It dominates perceptions of empirical descriptors or of autocorrelation descriptors from physical properties of atoms, for instance, of electronegativities, polarisations or volumes of chemical groups. The choice between these different methods is determined by the type of interactions of chemical compound with the receptor. By means of pattern recognition techniques you can make a classification of molecules on the base of the activity features. As well we obtaine fuzzy results with 75–95% probabilities only