Abstract: Inductively coupled plasma-mass spectrometry was used to determine the contents of 28 mineral elements in the stone and pulp of 73 durian samples from different cultivars grown in Malaysia, Thailand, Cambodia and Vietnam. The data obtained were analyzed by means of analysis of variance (ANOVA), principal component analysis (PCA), Fisher linear discriminant analysis (FLDA) and back propagation artificial neural network (BP-ANN) to develop and validate a model for discriminating durian from different geographical origins. The results showed that the contents of 16 and 13 mineral elements in the stone and pulp of durian significantly varied among growing areas, respectively. In PCA, the cumulative contribution rates of the first six principal components were higher than 85.207%, which could represent the major information about mineral element contents. The significantly differential elements were substituted into Fisher’s stepwise discrimination equation, and the results showed that the discrimination accuracy for single durian stone and pulp was low. However, combinations of mineral elements in durian stone and pulp significantly improved the discrimination accuracy. Through stepwise discrimination analysis, Li, Be, Mg, Mn and Rb in durian stone and Be, Ag and Ba in durian pulp were selected for modeling, and it turned out that the initial validation accuracy of the model was 91.8%, and the cross validation accuracy was 90.4%. The significantly differential elements were substituted into the BP-ANN model. As a result, As, Ag, Al and Rb in durian stone and Ag in pulp were selected as the top five most important elements for artificial neural network; the validation accuracy was 96.1% and 95.5% for the training and test sets, respectively. Our finding proved that it is feasible to distinguish durian from different southeast Asian countries by mineral fingerprint characteristics combined with chemometrics.

Key words: durian; mineral element fingerprint; geographical origin; traceability; chemometrics