A Real-Time 3D Reconstruction System for Screening Icosahedral Particles Under Different Conditions at the Microscope.
A Real-Time 3D Reconstruction System for Screening Icosahedral Particles Under Different Conditions at the Microscope.
复制标题
用于在显微镜下不同条件下筛选二十面体粒子的实时 3D 重建系统。
DOI：
10.1017/s1431927613005813
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
Baker,TimothyS
中科院分区：
文献类型：
--
作者：
Cardone,Giovanni;Yan,Xiaodong;Sinkovits,RobertS;Baker,TimothyS
自动数据收集系统[1]允许人们轻松地获取大量图像，然后可以通过单粒子技术处理这些图像以生成三维（3D）图。然而，确定这种高通量处理的最佳样品条件仍然需要漫长的手工工作。在某些情况下，有必要筛选不同的样品制备物以找到给出感兴趣的生理状态的样品制备物，而在其他情况下，筛选可用于优化缓冲液条件。这种分析可以通过使用允许在显微镜中加载多个网格的多样品架来促进，或者通过正在开发的将多个样品分配到单个网格上的新系统来促进[2]。尽管如此，每个条件的评估是基于对2D图像的分析，其仅提供关于样本的性质的部分信息。作为一个解决方案的分析粒子与二十面体对称性，我们正在开发一个实时的对齐和重建方案，提供了一个3D体积从一个单一的图像采集在显微镜下低温和低剂量的条件下。我们的新系统读取图像，因为它们是在显微镜采集和处理他们独立的一个小型计算机集群，使多达20个处理器并行使用。该实现提供了一个图形用户界面来监控结果，并且它依赖于AUTO3DEM包[3]进行所有处理步骤。用户需要的唯一输入参数是被检查颗粒的近似半径。每个微图最初都被合并以加速计算。然后并行执行散焦估计和粒子拾取。最后，所提取的粒子图像被用于估计3D密度图，而不使用任何参考模型。遵循随机模型方法[4]，使用随机方向生成十个初始模型，并且它们被独立地细化十次迭代，每次细化在2个处理器上计算。所有步骤均在无需用户干预的情况下执行。
Automated data collection systems [1] allow one to easily acquire large numbers of images that can be then processed by single-particle techniques to generate three-dimensional (3D) maps. However, the determination of the best sample conditions for such high-throughput processing can still require lengthy and manual work. In some cases it is necessary to screen different sample preparations to find the one giving the physiological state of interest, whereas in others screening can be used to optimize buffer conditions. This analysis can be facilitated by the use of multi-specimen holders that permit loading more than one grid in the microscope, or by novel systems under development that dispense multiple samples onto a single grid [2]. Still, the evaluation of each condition is based on the analysis of 2D images, which provide only partial information on the properties of the specimen. As a solution for the analysis of particles with icosahedral symmetry, we are developing a realtime alignment and reconstruction scheme that provides a 3D volume from a single image acquired at the microscope under cryogenic and low dose conditions.Our new system reads images as they are acquired at the microscope and processes them independently on a small computer cluster, making use of up to 20 processors in parallel. The implementation provides a graphical user interface for monitoring the results, and it relies on the package AUTO3DEM [3] for all processing steps. The only input parameter required by the user is an approximate radius of the particle being examined. Each micrograph is initially binned to speed up computations. Defocus estimation and particle picking are then performed in parallel. Finally, the extracted particle images are used to estimate a 3D density map, without using any reference model. Following the Random Model Method [4], ten initial models are generated using random orientations, and they are refined independently for ten iterations, with each refinement computed on 2 processors. All steps are performed without the need for user intervention.