Structure-camphor odour relationships using the Generation and Selection of Pertinent Descriptors approach

Structure-camphor odour relationships were carried out using GSPD (Generati on and Selection of Pertinent Descriptors) or GESDEM (Generation et Selecti on de Descripteurs et Elaboration de Motifs). This methodology consists in constructing descriptors from fragments. Their size (called order) is incre ased stepwise to obtain an optimum order giving a best classification. The set of studied compounds included 99 aliphatic alcohols, for which olfactor y properties have been described by Schnabel et al. [K.O. Schnabel, H.D. Be litz, C.V. Ranson, Untersuchungen zur Struktur-Aktivitats-Beziehung bei Ger uchsstoffen, Z. Lebensm.-Unters.-Forsch. 187 (1988) 215-223]. Classificatio n of compounds was tested by means of artificial neural network (NN) with b ack-propagation algorithm, Kth nearest neighbour (KNN) and discriminant ana lysis (DA). After GSPD, all compounds were well-classified and 93% of them were well-predicted by means of leave-one-out method using NN. A new test s ample composed of 42 alcohols (25 camphor and 17 non-camphor) with structur es analogue to these of training set was collected from Belstein and traine d on weight matrix of network which served in classification phase. The res ults of prediction were at 90.5% in agreement with those of literature. (C) 1999 Elsevier Science B.V. All rights reserved.