A theoretical example of one of these binary matrices is depicted in Table 1. From the individual sorting data collected, a symmetrical proximity matrix with sums of counts of how many times the attributes appear together in “sibling” relationships or “parent–child” relationships was compiled, similar to that in Table 2 but on a larger scale . To ensure that all data could be used to create the flavor wheel, 1st, the 2 groups were compared . Two separate similarity matrices were created, one for UC Davis participants and one for industry participants. The scaled matrices were used to run 2 separate 5-dimensional multidimensional scaling analyses . The results of the 5D-MDS analyses were used to run a multiple factor analysis , a technique to compare multiple datasets and in sensory science is typically applied to compare sensory profiles, also on XLSTAT 2015 . No significant difference was found between the 2 groups, so the data for all 72 participants were used for further analysis. To determine the clusters and levels of the flavor wheel, AHC was conducted on the similarity proximity matrix with cooccurrence values using the unweighted pair group average linkage agglomeration method . Hierarchical clustering is a statistical technique that can be applied to sorting data to group the attributes into different categories and subcategories on different levels in the form of a dendrogram . At the beginning of analysis, every individual object starts as a single “cluster,” and then the unweighted pair group average linkage links the attributes together, one pair at a time, from the bottom to the top . In each successive linkage, it merges the most similar pair of items . Upon observation of the dendrogram,grow bags garden truncation was set to specify 9 main categories. In addition to the analysis in XLStat, other methods of similarity were tested in R, such as Euclidean, maximum, and Manhattan. Other agglomeration techniques were tested in R as well, such as Ward’s, complete, single, and average .
In R, the Euclidean distance method with the unweighted average linkage method was determined to be the combination with the most distinct clusters without being biased by outliers or the size of clusters, but even this still split the Fruity group into 2. Otherwise, this combination was very similar to the XLStat result, confirming the hierarchical structure in 2 different software programs. Thus, the XLStat dendrogram with unweighted pair group average linkage, which kept the Fruity group intact, was selected. Finally, MDS analysis was performed to represent all 99 attributes in a 2-dimensional space, a visual aid to see where the attributes fell in proximity to one another.Nonmetric MDS was performed on the proximity matrix of Euclidean distance values, meaning the order of the “distance,” using Kruskal’s stress values, in the resemblance matrix matched the ranking of the distances for the representation space . MDS was performed to supplement the AHC data and to guide the positioning of the main classes around the new Coffee Taster’s Flavor Wheel. Since the similarity values were obtained from frequency counts for every pair of attributes, the data were considered nonmetric; that is to say, the differences or ratios between the values held no meaning. Higher values were considered more similar and lower values were considered less similar. Kruskal’s stress values, testing both Minkowski’s distance values and Euclidean distances commonly used in nonmetric MDS, were used to obtain the 2- in the stress function. In those results, the 9 main categories were dimensional coordinates that most closely adhere to the ranking positioned in the 2-dimensional space in a similar order, but the of those similarity values . Other plots were more sensitive to outliers, meaning some points were methods of MDS were tested in R, both metric and nonmetric, far from the origin of the MDS plot and the majority of points were clustered near the origin.
As the purpose of this analysis was to obtain the positioning of the main categories around the flavor wheel, the nonmetric MDS in XLStat was ultimately selected as the option that was less sensitive to outliers and most clearly separated the data points.The MFA comparison of the similarity matrices from the UC Davis panelist group and the coffee industry panelist group revealed that there was no significant difference between the 2 groups. The RV-coefficients were much greater than 0.70, meaning the 2 groups were related and came from the same population . An attribute-by-attribute comparison was also plotted from the MFA, showing the degree of similarity in sorting between the 2 groups for each attribute . For all participants together, AHC was truncated at 9 main classes, shown in 9 different colors . The MDS plot for the scaled data of all 72 participants is depicted in Figure 4. Using the dendrogram , the 9 main classes were named. Due to the fact that the lexicon was used to provide the attributes to be sorted, some main categories that were found did not have an “umbrella” term that existed in the lexicon, or a general word that encompasses and describes the category . In order to fit the AHC and MDS results onto a flavor wheel, a few modifications had to be made by SCAA and the researchers. Unfortunately, due to the nature of this project , it was impossible to know which of these “umbrella” terms would be needed exactly or how many, so a few of the terms were moved or added to the lexicon to create the final organization . This issue is further elaborated on in the Suggestions section. These 9 main categories are labeled in Figure 3 and 4. The attributes that are similar are found in the same categories and subcategories in the dendrogram . The attributes that are similar are found close to one another on the MDS plot and those that are less similar are further away from one another. Additionally, as mentioned earlier, the WCR Sensory Lexicon is a living document, so a few terms were added to the living WCR Sensory Lexicon document after the sorting exercise was complete, and as the lexicon was being finalized, based on the expert opinion of the scientists and panelists at Kansas State Univ.
Finally, with the unweighted pair group average linkage, there is a different similarity level for every single pair, and only 3 levels were needed for this flavor wheel. Thus, the dendrogram was interpreted by SCAA and the researchers to create a 2nd and 3rd tier of subcategories for each of the 9 main categories. To determine the positioning around the wheel, the MDS plot with category labels was used . Therefore, not only are the more similar attributes placed together in the same categories and subcategories, but the 9 main classes are placed around the flavor wheel based on similarity . The hierarchy used for the flavor wheel is the interpretation of the 9-class dendrogram in Figure 3 with these modifications incorporated. The final wheel, translated from Table 4, is depicted in Figure 5.The flavor wheel construction techniques used in this method created a suitable, intuitive flavor wheel to complement the Sensory Lexicon for the specialty coffee industry. However, there are ways to improve the process from the beginning if these methods are to be adopted for the construction of flavor wheels for other products. If the researchers know that a product lexicon will be used to develop a wheel or other visual containing multiple categories and tiers, then these projects could be coordinated to improve the process. To begin with, the initial lexicon should contain only vocabulary from the most specific attributes . The study subjects would then be able to use a free sorting exercise similar to that performed in the study, but the exercise would not involve multiple levels. The subjects would simply sort the words into as many groups or clusters as they deem necessary. Also, when the descriptors are presented to the subject to be sorted, it would be best to randomize them for each individual,grow bag for tomato rather than presenting the same unorganized lexicon to each subject. In this way, both research projects would inform each other as they progressed. After the initial sorting exercise, a cluster analysis and MDS analysis could be performed to determine the number of groups for the 2nd tier and the positioning of the words around the wheel, respectively. These 2nd-tier clusters would then be appropriately named by the subjects or descriptive panel in a consensus exercise. Next, the sorting exercise would be repeated with only the 2nd-tier vocabulary, to sort those descriptors into clusters. Finally, the 1st-tier groups would be named, with input from the descriptive panel. To summarize, to use this improved flavor wheel construction technique, researchers would develop the lexicon and wheel simultaneously. Only the most specific vocabulary words should be present in the initial lexicon, and then the more general descriptors, or so-called “umbrella” terms, would be added in later, with help from the descriptive panelists for as many iterations or levels deemed necessary.There are multiple ways in which removal of infected host plant tissue can be employed as an element of disease management. These include removal of reservoir hosts to limit pathogen spillover onto a focal host , roguing of infected focal hosts to limit secondary spread , and removal of localized infections within hosts to limit further infection or to retrain an unproductive plant . Studies of bacterial pathogens in perennial crops have evaluated the utility of pruning as a disease management tool, with mixed results . The removal of infected plant tissues is analogous to measures used for management of trunk diseases, often referred to as “remedial surgery,” as an alternative to replacing infected plants . In this study, we investigated whether severe pruning of Xylella fastidiosa-infected grapevines in commercial vineyards could clear vines of existing infections. Pierce’s disease is a lethal vector-borne disease of grapevines caused by the bacterium X. fastidiosa . After susceptible plants are inoculated by X. fastidiosa, pathogen populations multiply and move through the xylem network, leading to symptoms of reduced water flow , including leaf scorch, cluster desiccation, vine die back, and eventually death.
There is no cure for grapevines infected with this bacterium; current strategies for management of PD in California vineyards involve limiting pathogen spread to uninfected vines by controlling vector populations, disrupting transmission opportunities, and eliminating pathogen sources in the surrounding landscape . PD is notable for the numerous sources of variability in infection levels and symptom severity in plants. X. fastidiosa infection levels vary among plant species , grapevine cultivars , seasons , and as a function of temperature . Like other bacterial plant pathogens , X. fastidiosa is often irregularly distributed within individual hosts. For example, X. fastidiosa infection levels in grapevines may vary by more than 10-fold between grapevine petioles and stems ; in other hosts, infection levels may vary by more than 100-fold between basal and apical sections of shoots . This within-host heterogeneity may be epidemiologically significant if it affects pathogen acquisition efficiency . Moreover, if such variation is associated with protracted localized infection near inoculation points, such heterogeneity may facilitate other disease management tactics. In addition to grapevines, other plant species that are susceptible to X. fastidiosa infection include citrus in South America . Management of the resulting disease in C. sinensis relies on clean nursery stock, vector control, and pruning infected plant tissue from established trees or roguing young plants . The concept of pruning of infected plant material is based on the fact that, in established trees , tissue with early symptoms of infection can be pruned ~1 m proximal to the most symptomatic basal leaf, effectively eliminating infections, as the remaining tissue is free of X. fastidiosa . However, pruning is not adequate for young trees or for removing bacterial infections if any symptoms are present in fruit . X. fastidiosa multiplies and spreads through the xylem vessels, reaching the roots of perennial hosts such as citrus , peach , alfalfa , and blueberry . Nonetheless, under field conditions, chronic infection of grapevines is temperature and season dependent. In regions with freezing winter temperatures, infected plants can recover in winter, curing previously infected and symptomatic grapevines .