Our goal is to develop the software technology required to use electron microscopy to determine the atomic structures of isolated macromolecular assemblies in a routine and rapid fashion. Images of a very large number of isolated (single) macromolecular particles are needed in order to achieve the signal-to-noise ratio at high resolution that is required for this goal. A data set of approximately 100,000 asymmetric units can produce a three-dimensional reconstruction at 8-12Angstroms resolution. This resolution is sufficient to insert, and adjust if necessary, atomic-resolution models of component macromolecules, whose structure would have been determined previously by other methods (X-ray or electron crystallography, or NMR spectroscopy). Even larger data sets, consisting of at least one million asymmetric units, are needed to achieve resolutions better than 5Angstroms, and ultimately to obtain 3-D reconstructions at about 3.5Angstroms, a level of resolution which is needed for de novo determination of an atomic structure. With data sets of this size, new software tools must be created in order to complete the computational work in a rapid fashion. The steps that we have identified for work within this Program Project are: (1) Single-particle software will be optimized to run on highly parallel computers, since processing of such large data sets on a single-node workstation becomes unrealistically long. (2) Computer-assisted """"""""boxing"""""""" (identification) of single particles will be improved to such an extent that tedious, human editing of galleries of candidate particles is no longer necessary. (3) Better algorithms will be developed to estimate the alignment-parameters of single particles. In addition, the Program Project will include key elements of core research and infrastructure. Experimental images will be obtained for two macromolecular assemblies whose atomic structures are already known. These reference-data are needed for the development and validation of the new software technology that the Program will create.
Showing the most recent 10 out of 22 publications
