Figure 4.

Two main data clusters can be identified by inspection (the a = 111 Å group and the a = 114–115 Å group). We observed that our data partitioned cleanly between 28 data sets with an a (= b) unit-cell parameter of approximately 114–115 Å and 118 data sets with an a (= b) unit-cell parameter of approximately 111 Å. The separation into the two unit-cell clusters is shown in the monochrome clustering on the left. The further division of these two clusters into amplitude-based clusters is shown by the colors on the right. The a = 114–115 Å unit-cell-based cluster contains the green and red clusters and the a = 111 Å unit-cell-based cluster contains the cyan, purple and yellow clusters. Each of our data sets was sufficiently large that amplitude-based clustering could have been used from the start. However, many serial crystallography projects consist of narrow wedges of data, each of which might be too small to cluster effectively using amplitudes because amplitude-based clustering requires that data sets have a sufficient number of observations in common. This figure illustrates how an initial use of cell-based clustering might be used to bootstrap amplitude-based clustering.