BLEND: Management of Data from Multiple Crystals

Among the strategies designed to avoid data corruption caused by radiation damage, collection from multiple crystals is already playing a major role. The idea is to shoot many crystals of a same structure recording only partial, essentially radiation damage free, data sets from each. Subsequent collation of the different partial datasets together could lead to the formation of a complete dataset, if the different crystals possess a good degree of isomorphism. Aside from the advantage of limiting radiation damage, the use of multiple crystals brings with it a potential increase in data multiplicity; this, in turn, has beneficial effects on the amplification of the anomalous signal to noise ratio for phasing, as recently exploited by Hendrickson (Science 336, 1033 (2012)).

BLEND is a new program to manage data collected from multiple crystals. The required input consists of two or more integrated datasets. The program, then, carries out cluster analysis to identify groups of crystals that are most similar to one another. The user can, at this point, select one or more groups from the cluster analysis dendrogram, (Figure 1) and run BLEND again to merge and scale all groups in the selection. The program also checks individual datasets for overall radiation damage and eliminates images whose intensities are deemed to be reduced below a given limit. Estimates for resolution cutoffs are also provided through the use of spline interpolation of integrated intensity averages.

Figure 1. All nodes at which two or more datasets merge lead potentially to higher completeness, higher redundancy and good merging statistics.

Several tests on BLEND have shown that of the thousands of combinations implied by the combinatorial explosion from multiple crystals, those suggested by the cluster analysis carried out in the program rank among the best, in terms of merging statistics like Rmeas and Rpim (Figure 2). In this respect it could be stated that BLEND automatically performs an optimal selection of multiple crystals.

Figure 2: In this plot, the Rmeas spread of values deriving from the random combinations of 2, 3, 4, ... datasets is displayed. Full lines join the 75th and 25th percentiles for the spread, while the broken line is the median. The open circles correspond to combinations selected by cluster analysis in BLEND, the majority of which fall below the 25th percentile, thus representing selections with close to optimal merging statistics.

For more information on BLEND please contact us on industry@diamond.ac.uk.