The goal of this project is to fill in the gap between research developments in cryo-electron microscopy (cryo-EM) and recent advances in large-scale scientific computing, and thus to better enable high-resolution single particle reconstructions on a routine and timely basis. To accomplish this goal, we will conduct research in the design and implementation of new numerical algorithms that will lead to significant reduction in the amount of work required to perform high-resolution single particle reconstructions. Our work will include improving the efficiency of particle orientation search by developing a parallel algorithm that performs progressively localized projection matching, developing techniques for enhancing the convergence and robustness of 3-D reconstruction, seeking alternative local optimization algorithms to simultaneously correct both the particle orientation parameters and the 3-D structure, and combining the local optimization scheme with a global combinatorial search strategy to be developed in Project E. In addition to designing and implementing new algorithms, we will also develop strategies for highly parallel computation to speed up the compute-intensive tasks in single particle reconstructions. These will include optimal data decomposition, task scheduling, and techniques for exploiting memory hierarchy and reducing I/O latency. In all of the areas of this research, we will interact and collaborate closely with investigators from the other projects in this Program, the primary goal being to ensure that the results of our work are integrated into SPIDER (Project A), SPARX (Project B), software technology that is developed for automated boxing of images of single particles (Project D), and software technology that is developed for global optimization of inter-particle alignment parameters (Project E).
Showing the most recent 10 out of 22 publications
