High Resolution in Single Particle Reconstruction

描述（由申请人提供）：我们描述了一项研究计划，旨在开发用于单粒子分析的扩展图像形成模型和伴随的计算方法，目标是在生物大分子的电子显微镜结构研究中实现近原子分辨率。虽然在一些初步研究中已经有可能获得比10a更细的分辨率，但仍然没有常规高分辨率电子显微镜结构测定的计算工具。我们的目标是通过开发鲁棒和精确的图像处理和误差分析算法，提高现有的能力，以便从单个粒子的固有噪声图像中提取4- 7a分辨率的3-D结构。我们首先证明了目前的电子显微镜图像形成模型与实验证据相矛盾，其次，使用现有的图像处理工具很难实现高分辨率结构确定所需的大量数据处理。在本提案的框架内，我们将详细说明并实验测试电子显微镜的图像形成模型，该模型考虑了一些非线性效应。具体来说，我们建议应该有一个依赖于信号的噪声，因此它将取决于存在于网格上的蛋白质的相对数量。该模型将为电磁数据的信噪比估计和对准方法的改进提供依据。为了减少数据量，我们将开发一个高度精确的算法库，用于从投影重建三维图像，并使用新颖的插值技术进行二维图像处理。我们还将开发用于计算从其投影集重建的结构中的实际空间方差/协方差的算法。这些方法旨在利用冷冻电镜定位构象变异性，并提高x射线晶体学域与三维电子显微镜图对接的准确性。该软件将以确保完全可移植性的方式开发，并将在领先的软件包SPARX和SPIDER中在整个新兴市场社区传播。

DESCRIPTION (provided by applicant): We describe a research plan to develop an extended image formation model for single particle analysis and accompanying computational methods with the goal of achieving near atomic resolution in the electron microscopy structural studies of biological macromolecules. Although it has been possible to attain resolution finer than 10 A in a number of pilot studies, there are still no computational tools for routine high-resolution electron microscopy structure determination. Our goal is to advance the existing capabilities in order to extract 4-7 A resolution 3-D structures from the inherently noisy images of single particles by developing algorithms for robust and accurate image processing and error analysis. We demonstrate first that the current image formation model in electron microscopy is contradicted by the experimental evidence, and secondly that the massive data processing required for the high-resolution structure determination will be difficult to achieve using existing image processing tools. Within the framework of this proposal, we will detail and experimentally test an image formation model of electron microscopy that accounts for some non-linear effects. Specifically, we suggest that there should be a signal dependent noise, which therefore will depend on the relative amount of protein present on the grid. The new model will provide a basis for the signal-to-noise estimation for EM data and for improvements of alignment methods. In order to reduce the volume of the data we will develop a library of highly accurate algorithms for 3-D reconstruction from projections and for 2-D image manipulation using novel interpolation techniques. We will also develop algorithms for the calculation of the real space variance/covariance in structures reconstructed from sets of their projections. These methods are aimed at the localization of conformational variability using cryo-EM and at improving the accuracy of the docking of X-ray crystallographic domains into 3-D EM maps. The software will be developed in ways that assure full portability and will be disseminated throughout the EM community within the leading software packages SPARX and SPIDER.