Figure 4. Cluster analyses of 22 CHs in LD and DD.
Two independent fitted values per treatment (FA-normalized mean and FA-normalized slope βj) are used. Clusters based on mean data are shown in A-D, clusters based on slope data, E–F. A,B: MDS/AP clusters of FA-mean data in LD and DD. X and Y axes are multidimensional scaling projections of the 22-dimensional correlation distance matrix into two dimensions. Compounds are joined by a line to a central “exemplar” cluster member if the Affinity Propagation clustering algorithm of Frey and Dueck places them in the same cluster. Different clusters may overlap in this 2-D representation, as two dimensions is not sufficient for MDS projection to capture all of the variation in the data; the AP clusters are determined from the full 22-dimensional structure, which has 4 significant principal components. C,D: Ward's minimum variance clustering of wild-type FA-mean data in DD and LD. E,F: MDS/AP clusters of FA-slope values βj in LD and DD.