Application of classification-tree models to characterize the mycobiota of grapes on the basis of origin

Classification-tree (CT) models are a simple and robust exploratory data analysis technique that can be used in classification, regressions and summaries of data. They distill complex ecological relationships into simplified rules and identify the species necessary for sample classification on the basis of detailed ecological inventories. The usefulness of this technique to characterize and represent differences in the grape mycobiota of distinct origins was evaluated. Grapes from four Portuguese winemaking regions were selected for a 3-year study: Alentejo, Douro, Ribatejo and Vinhos Verdes. The mycobiota of grapes was assessed with plating methods and the frequencies of isolations of the fungal taxa identified in 32 samples were used as a training dataset. The CT algorithm selected the fungal taxa and respective frequencies to classify grapes according to its region of origin. The ten-fold cross-validation technique was used for model evaluation. The success rate of the model was quantified and expressed in the number of correctly classified samples overall and into region. Furthermore, model refinement was performed using attribute selection algorithms and class redefinition. A simple tree model was generated that classified grapes into three regional origins: Douro, South (Alentejo and Ribatejo classes together) and Vinhos Verdes, on the basis of the incidence of Aspergillus niger aggregate and Penicillium thomii in samples with an accuracy of 82%. The merits and demerits of these models are discussed.