Figure 6.

Amplitude-based clusters generated using KAMO (dendrogram). This dendrogram shows a representation of the similarity of pairs of data sets and of clusters of more data sets. They are arranged with the most similar clusters near each other and the connecting bar at a height corresponding to the distance between clusters. The difference was calculated using Ward’s method for hierarchical clustering, which yields a composite metric that contains information from amplitude differences and from unit-cell differences. Our algorithm is described in Section 2.3. Structures were solved corresponding to each of these 145 clusters. We deposited the structure derived from refinement against structure factors, each of which was an average of that observation from all of these clusters, as PDB entry 7kty. Through inspection of the derived structures we selected a height within the dendrogram at which to partition our data, giving five clusters. We then averaged all structure factors within each of the five distinct clusters and then refined against these to give cluster-average structures. We deposited the averaged structure from the green clusters as PDB entry 7ku1, from the red clusters as PDB entry 7ku2, from the cyan clusters as PDB entry 7ku3, from the purple clusters as PDB entry 7ktz and from the yellow clusters as PDB entry 7ku0.