PARTICLE ESTIMATION FOR ELECTRON TOMOGRAPHY

这个子项目是许多研究子项目中的一个 由NIH/NCRR资助的中心赠款提供的资源。子项目和 研究者（PI）可能从另一个NIH来源获得了主要资金， 因此可以在其他CRISP条目中表示。所列机构为 研究中心，而研究中心不一定是研究者所在的机构。 塑料切片的电子层析成像（ET）在质量上受到数据低信噪比（SNR）的限制。 这个问题对于冷冻水合样品的ET（cryo ET）甚至更大，因为它们具有低对比度并且对电子束的损伤非常敏感。 所有EM样本的SNR通常是试图实现的分辨率的递减函数，严重限制了断层图像中可访问的结构细节的量。 对于包含多个副本的给定结构的标本，我们已经开发了一种算法，以提高信噪比，通过估计真实的三维结构，存在于细胞中，基于图像的多个副本的同一结构，往往是可见的断层图像。 我们的技术建立在为单粒子电子显微镜开发的方法的基础上，主要的区别是对齐和平均发生在3D断层体积上。这种方法的优点是可以原位研究感兴趣的结构，而不必从其细胞环境中分离出来。 我们用于估计真实3D颗粒结构的算法如下。 使用断层图像内手动选择的颗粒位置，切除包含每个颗粒的子体积，然后通过在离散欧拉旋转的范围内明确地将每个子体积与参考体积进行比较来旋转地对准。 子体积比较通常使用傅立叶域局部相关系数序列函数来计算，该函数还为每次旋转提供最佳平移移位。 参考体积可以从粒子的集合中选择，或者可以通过粒子子集的成对二叉树比对来生成无偏参考。 该对准过程通常是迭代的，从而减小旋转搜索空间和粒度，并且允许在每次迭代时更新参考体积。 一旦我们对每个粒子进行了旋转和平移估计，我们就通过对对齐的子体积进行平均来估计粒子体积。 我们通过考虑傅立叶分量贡献或缺乏傅立叶分量贡献来补偿缺失数据的楔形，这是单轴倾斜ET的特征，因为每个粒子被转换为与参考对准。 定性地，我们的粒子估计算法允许我们可视化在原始断层图像中不可见的结构细节。 定量地，频谱信噪比测量显示SNR改善接近于平均粒子数的预期。 在最近的过去，我们已经增加了2个功能，这个软件：平均粒子从多个断层图像的能力，并能够分布在计算机网络的最佳对齐搜索。 实验表明，在最终平均值中更好地考虑了缺失的楔形，结果可以添加具有更多取向变化的颗粒。计算时间也从几天缩短到几个小时。

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Electron tomography (ET) of plastic sections is limited in quality by the low signal-to-noise ratio (SNR) of the data. This problem is even greater for ET of frozen-hydrated samples (cryo ET) because they have low contrast and are very sensitive to damage by the electron beam. SNR for all EM samples is typically a decreasing function of resolution one is trying to achieve, severely limiting the amount of structural detail that is accessible in a tomogram. For specimens containing multiple copies of a given structure, we have developed an algorithm to improve the SNR by estimating the true 3D structure that is present in the cell, based on the images of multiple copies of the same structure that are often visible in a tomogram. Our technique builds on the approach developed for single-particle electron microscopy, with the primary difference that alignment and averaging occur over the 3D tomographic volume. The advantage of this approach is that the structure of interest can be studied in situ instead of having to be isolated from its cellular context. Our algorithm for estimating the true 3D particle structure is as follows. Using manually selected particle locations within the tomogram, a sub volume containing each particle is excised and then aligned rotationally by explicitly comparing each sub volume with a reference volume over a range of discrete Euler rotations. Sub volume comparison is typically computed using a Fourier domain local correlation coefficient sequence function, which also provides the optimal translational shift for each rotation. The reference volume can be chosen from the collection of particles, or an unbiased reference can be generated by a pair-wise binary tree alignment of a subset of particles. This alignment procedure is typically iterated, reducing the rotational search space and granularity, and allowing an update of the reference volume at each iteration. Once we have rotation and translation estimates for each particle we estimate the particle volume by averaging the aligned sub volumes. We compensate for the wedge of missing data that is characteristic of single-axis tilting ET by accounting for the Fourier component contribution, or lack thereof, from each particle as it is transformed into alignment with the reference. Qualitatively, our particle estimation algorithm allows us to visualize structural details that are not visible in the original tomogram. Quantitatively, the spectral-signal-to-noise ratio measurements show SNR improvements close to that expected for the number of particles averaged. During the recent past, we have added 2 functions to this software: the ability to average particles from multiple tomograms and the ability to distribute the search for optimal alignment across a network of computers. Experiments have shown that the missing wedge is better accounted for in the final average with the result that particles with more orientational variations can be added. The computation time has also been cut from a few days to a few hours.