NEURAL NETS FOR THE SIMULATION OF MOLECULAR RECOGNITION WITHIN MS-WINDOWS ENVIRONMENT

A neural method for the comparison of the molecule series, using a sel f-organizing map (SOM) procedure, is presented. The coordinates of ind ividual atoms within the most active molecule mimicking the geometry o f the receptor are trained with SOM algorithms to form a template whic h provides a basis for recognizing the similarity within the compounds series. Therefore the SOM matrices produced for individual molecules, if analyzed further by neural procedures or by the calculation of the matrix determinants, provide reasonable classifications according to the molecule's structure or actual activity observed, provided that th is is limited by shape factors. All computational operations are perfo rmed within MS-WINDOWS environment, using commercial programs of HYPER CHEM and MATLAB.