The application of hierarchical cluster analysis to the selection of isomorphous crystals.

It is generally assumed that the quality of X-ray diffraction data can be improved by merging data sets from several crystals. However, this effect is only valid if the data sets used are from crystals that are structurally identical. It is found that frozen macromolecular crystals very often have relatively low structure identity (and are therefore not isomorphous); thus, to obtain a real gain from multi-crystal data sets one needs to make an appropriate selection of structurally similar crystals. The application of hierarchical cluster analysis, based on the matrix of the correlation coefficient between scaled intensities, is proposed for the identification of isomorphous data sets. Multi-crystal single-wavelength anomalous dispersion data sets from four different protein molecules have been probed to test the applicability of this method. The use of hierarchical cluster analysis permitted the selection of batches of data sets which when merged together significantly improved the crystallographic indicators of the merged data and allowed solution of the structure.