Complete linkage clustering

Twenty populations were sampled with plants from Hailuoto, at 65°00 N, representing the northernmost site and material from Hanko, at 59°49 N, representing the southernmost collection site (and, incidentally, the southernmost point of land in the country). Five samples represented central Finland with the remainder originating from the southern part. Fifty-five compounds were detected by GC-MS analysis, 53 of which were identified. The data obtained were subjected to complete linkage analysis, which differentiated several clusters that corresponded moderately well with geography. Genetic distance values derived from the RAPD data correlated well with chemical distance values determined from the terpene data (r=0.41, P<0.0001). 

Figure 2.13 Vhe use of cluster analysis to display the sisilarity of stationary phases with the complete linkage furthest neighbor dendrogras.

There exist several methods of hierarchical clustering which use diverse measures of distance or similarity, respectively, e.g., single linkage, complete linkage, average linkage, centroid linkage, and Ward s method (Sharaf et al. [1986], Massart et al. [1988], Otto [1998] Danzer et al. [2001]). 

FIGURE 3.24 Dendrogram of fatty acid concentration data from mummies and reference samples. Hierarchical cluster analysis (complete linkage) with Euclidean distances has been applied. 

Compute the distances between all clusters using complete linkage, single linkage, average linkage, or other methods. 

FIGURE 6.11 Resulting hierarchy from complete linkage applied to the data of Figure 6.10. Thicker lines correspond to higher levels in the hierarchy, and thus to larger clusters. 

Figure 6. Heat map of a small cluster of related antihistaminic drugs. 149 pharmacological assays are clustered on the X-axis and 10 compounds are clustered on the Y-axis. Clustering is performed using Pearson correlation and complete linkage using a pICso data set. For nonhits with primary screening % inhibition values of less than 20%, a default pICso value of 3.5 was used. Non-hits and above 20% inhibition received a default pICso value of 4.0. The pICso values range from the default value of 3.5 (blue-green) to 9 (red). Black spaces indicate data that is missing due to compound interference with the detection method.

Here too, Pearson clustering with complete linkage can be applied to identify compounds with similar ADME profiles. Having identified BioPrint drugs with similar ADME profiles, the BioPrint pharmacokinetics database (which contains literature pharmacokinetic data on over 1000 drugs) is queried and predictions for the test compound are made based on the pharmacokinetic profile of the ADME nearest neighbors. 

Complete linkage, furthest neighbor - uses the greatest distance between any two samples in the different clusters. 

Complete linkage 0.5 0.5 0 0.5 max(diUk + da k) Many small clusters... 

Figure 7.1 Authentication of monovarietal virgin olive oils results of applying clustering analysis to volatile compounds. The Mahattan (city block) distance metric and Ward s amalgamation methods were used in (a) the Squared Euclidean distance and (b) complete linkage amalgamation methods. Note A, cv. Arbequina (6) C, cv. Coratina (6) K, cv. Koroneiki (6) P, cv. Picual (6) 1, harvest 1991 2, harvest 1992. Olives were harvested at three levels of maturity (unripe, normal, overripe) (source SEXIA Group-Instituto de la Grasa, Seville, Spain).

In the first complete linkage analysis, the interaction term (layers, 1/1, float) was assigned the same weight as any of the other relevant attributes. The algorithm yielded three clusters (Figure 3) when there was only one match (or equivalently, when the maximum distance between clusters was 12 no matches ). Two of these clusters accounted for only 13 of the 101 location points. Interestingly, 11 of these 13 points were precisely those and only those points that revealed no pseudomorphs. The other two points were those for which pseudomorphic evidence could not be identified. As such, these two small clusters were taken together to form the cluster of inde-terminates . When matches were increased to three, the big cluster of 88 location points was subdivided into three smaller clusters. A cluster could represent either a fabric type or a fabric pattern. 

Figure 3. Complete linkage cluster analysis unweighted interaction.

Allen, F. H., Doyle, M. J., and Taylor, R. Automated conformational analysis from crystallographic data. 2. Symmetry-modified Jarvis-Patrick and complete-linkage clustering algorithms for three-dimensional pattern recognition. Acta Cryst. B47, 41-49 (1991). 

---