STRUCTURE MUSK ODOR RELATIONSHIPS FOR TETRALINS AND INDANS USING NEURAL NETWORKS (ON THE CONTRIBUTION OF DESCRIPTORS TO THE CLASSIFICATION)

Structure-musk odor relationships were established by means of a 3-layer neural network (NN) using the back-propagation algorithm. To test the reliability of the NN approach, structure-odor relationships were established for a set of 53 tetralins and for subsets of 45 and 41 tetralins, and tested using a set of 15 indans and the remaining tetralins (8 or 12 compounds) as test sets. Each molecule of the training set was described by 7 variables coding substituents of 6 free sites of the tetralin or indan rings (6 steric hindrance descriptors and one electronegativity descriptor). Odor was coded by a binary variable. Training the NN with 53 tetralins gave 100% correct classification and 100% good prediction for the test with indans in all the trials. With the subsets of 45 and 41 tetralins, the classification ability of the NN was in all cases higher than 96.4% and its prediction ability higher than 92.6%. The contributions of the descriptor variables to the classification were evaluated according to different methods. The obtained results confirm the well-known effects of the steric hindrance of the functional group and of substituents in the ortho position.