Classification of images of biomolecular assemblies: a study of ribosomes and ribosomal subunits of Escherichia coli.

Images of macromolecules obtained in the electron microscope are subjected to correspondence analysis. The structure inherent in the data in the resulting low-dimensional factor space is characterized by a mixed classification method which combines the dynamic clouds clustering technique with hierarchical ascendant classification (HAC). For our data, the rejection of marginal clusters obtained by dynamic clouds clustering appears as a crucial prerequisite for a stable performance of HAC. The method is applied to two sets of 204 and 177 images that show the 70S ribosome of Escherichia coli, in the range of overlap views as defined by A. Verschoor and co-workers, and to two sets of 480 and 496 images of the 50S subunit of E. coli depleted of L7/L12 proteins in the well-defined crown view. Reproducible classes are obtained, which are characterized by images reconstituted from factorial coordinates. These classes appear to be related to different orientations on the specimen grid (in the case of the 70S particle) and to different conformational states (50S subunit).