Development of structure-taste relationships for monosubstituted phenylsulfamate sweeteners using classification and...

Twenty monosubstituted phenylsulfamates (cyclamates) have been synthesized and have had their taste portfolios determined. These have been combined with 63 compounds already in the literature to give a database of 83 ortho, meta, and para compounds. A training set of 75 compounds was randomly selected leaving eight compounds as a test set. A series of nine predictors determined with Corey-Pauling-Koltun models, calculated from the PC SPARTAN PRO program and Hammett sigma values taken mainly from the literature, have been used to establish structure-taste relationships for these types of sweeteners. The taste panel data for all compounds were categorized into three classes, namely, sweet (S), nonsweet (N), and sweet/nonsweet (N/S), and a novel "sweetness value" or weighting was also calculated for each compound. Linear and quadratic discriminant analysis were first used with the S, N, and N/S data, but the results were somewhat disappointing. Classification and regression tree analysis using the sweetness values for all 75 compounds was more successful, and only 14 were misclassified and six of the eight test set compounds were correctly classified. For the 29 meta compounds, one subset using just two parameters classified 83% of these compounds. Finally, using various methods, predictions were made on the likely tastes of a number of meta compounds and a striking agreement was found between the tree prediction and those given by earlier models. This appears to offer a strong vindication of the tree approach.