Electron tomography of complex biological specimens on the Grid

Electron tomography allows elucidation of the three-dimensional structure of large complex biological specimens at molecular resolution. To achieve such resolution levels, sophisticated algorithms for tomographic reconstruction are needed. Iterative algebraic algorithms yield high quality reconstructions, but they are computationally expensive and high performance techniques are needed to exploit them in practice. We present here a grid computing approach for tomographic reconstruction of large biological specimens. The approach is based on the computational Single-Program-Multiple-Data model, which basically decomposes the global problem into a number of independent 3D reconstruction subproblems. New performance metrics and job submission policies are proposed here that could be of general interest in the field of Grid Computing. We have evaluated this approach on the grid hosted by the European EGEE (Enabling Grids for E-sciencE) project. The influence of the problem size and the parallelism grain has been thoroughly analyzed. Our results demonstrate that the grid is better suited for large reconstructions, as currently needed in electron tomography. To fully exploit the potential of computational grids, the global problem should be decomposed into an adequate number of subdomains.