STRUCTURE-ODOR RELATIONSHIPS - USING NEURAL NETWORKS IN THE ESTIMATION OF CAMPHORACEOUS OR FRUITY ODORS AND OLFACTORY THRESHOLDS OF ALIPHATIC-ALCOHOLS

Structure-odor relationships were established for a sample of 99 aliph atic alcohols using a three-layer backpropagation neural network. The molecular structure was described using a common skeleton with six pos sible substitutions. Substituents were described using only their van der Waals volumes. The discrimination between fruity and camphoraceous odors of 67 compounds gave good results in classification (100%) and prediction (85%) phases. With the global set, the network correctly cl assified and predicted the camphoraceous character of compounds (100% and 95% respectively) but gave poorer results for the fruity character (87% and 74% respectively). Calculations of pOLs (pOL = -log (olfacto ry threshold expressed in mol/L)) of 45 camphoraceous compounds were a lso made. When all camphoraceous compounds were used to establish the model, 91% of the pOLs were correctly estimated. When attempts were ma de to predict the pOL values of 10% of the compounds from a model desi gned using 90% of the sample, only 74% of the pOLs were correctly esti mated.