ARCIMBOLDO-SX
Final Report Abstract
During this first year of the project several algorithms and tools have been programmed and tested, to be eventually incorporated in the supercomputer version of ARCIMBOLDO-SX. One of these tools enables direct methods to be applied to difficult cases where massive computational power may be needed to render a solution. This tool tries an ab initio solution with Patterson methods for a native structure or a heavy atom substructure from derivative data with SHELXD and further density modification and phase expansion with SHELXE incorporating the native data. For some macromolecular structures one single ARCIMBOLDO run can create ten thousands of submitted jobs which even on a supercomputer grid is quite a challenge to CPU and memory and occupies a lot of disk space. Therefore, an analysis of figures of merit at various stages was established, in order to prioritize the more promising partial solutions, terminate hopeless ones and in general, manage the ARCIMBOLDO-SX runs to optimize resources. The output from an ARCIMBOLDO run is evaluated to backtrace the candidates for possible solutions and to enable a preselection of solutions based on specific numbers from the Phaser output. If only those preselected solutions are sent to further processing with SHELXE, i.e. not all candidates are expanded, the computing time thus saved can be exploited to improve the starting phases prior to expansion, implement modeling and test derived hypotheses in parallel. Selecting a few most promising partial solutions to be pushed in a sort of “express lane” allows in its turn to spend more computational resources in sophisticated procedures that would be unmanageable on a larger scale. Structure building on the mainchain traces outcoming from SHELXE requires an ideal resolution of about 2Å. In cases of significant deviation from this value density modification and autotracing of partial ARCIMBOLDO solutions becomes less effective. To overcome this resolution limitation the starting map prior to density modification was improved with the program BUSTER which generates Maximum Likelihood enhanced phase distributions encoded as Hendrickson-Lattman coefficients. Scwrl4 was then applied to generate all possible matches of the mainchain trace from SHELXE with the sequence respecting the secondary structure prediction and general structural validation criteria. Those results were evaluated through further refinement with BUSTER. In an iterative way, this method allows automatic solution and building of macromolecules and has been successful in the solution of a previously unknown protein structure at 2.7Å and a protein–DNA complex at 2.8Å.
Publications
- "High-Resolution Crystal Structure of MltE, an Outer Membrane-Anchored Endolytic Peptidoglycan Lytic Transglycosylase from Escherichia coli", Biochemistry, 2011, 50, 2384–2386
C. Artola-Recolons, C. Carrasco-López, L. I. Llarrull, M. Kumarasiri, E. Lastochkin, I. Martínez de Ilarduya, K. Meindl, I. Usón, S. Mobashery, J. A. Hermoso
