Compute Cluster for in vitro and in situ Analysis of Molecular Machines

项目概要/摘要 密歇根大学（U-M）建立了一个尖端的冷冻电子显微镜（cryo-EM）设施 配备最新直接电子探测器（DED）的多台高分辨率显微镜。的 该设施使用自动数据采集和简化的图像处理协议，其中一些是在密歇根大学开发的， 其允许使用单颗粒和 冷冻电子断层扫描（cryo-ET）策略。改进的摄像头和自动化数据的结合 收集现在使用户快速收集确定亚3.5微米结构所需的数据成为常规， 分辨率，允许在所得密度图中直接建立模型。然而，每个冷冻EM数据集 需要数TB的存储空间和数千个CPU（中央处理器）或GPU（图形处理器 单位）小时的计算机系统配备了大量的随机存取存储器（RAM）。这使得 高通量数据收集是低温EM管道中的新瓶颈。有效存储的能力， 处理和分析日益庞大的数据集对于这一过程至关重要。因此，能够快速 收集大型数据集，再加上迅速扩大的校园用户群，意味着访问 计算资源现在代表了许多U-M冷冻EM项目成功的主要限制因素。 密歇根大学的冷冻EM计算资源，位于生命科学研究所 (LSI)，最初是通过S10赠款的一部分，以大学，并将达到他们的生命结束， 2021年5月，他们将不再得到供应商的支持。这项建议要求提供资金， 老化的计算和存储系统，以及集成一个强大的计算集群， 有足够的内存来有效地确定大颗粒的原子分辨率结构（> 1，000像素框大小）。 这个升级的计算集群将包括1，248个英特尔CPU核心，16个NVIDIA Quadro RTX 8000 GPU （图形处理单元）、两个大型内存节点和820 TB存储空间，这些将被合并， 与我们的一些较新的CPU/GPU计算节点集成，这些节点将保持在线。安装此 资源是必不可少的：1）密歇根大学低温EM设施的能力，以提高数据收集的四个高分辨率 具有“即时”处理的冷冻电子显微镜; 2）大型异质冷冻电子显微镜的图像处理 数据集（+1，000，000个粒子）;以及3）具有足够内存以容纳结构化 分析大的大分子复合物，如病毒和细菌分泌系统，以及大的 体积用于原位冷冻ET分析。这里提出的计算基础设施将大大 提高国家的最先进的冷冻EM结构确定在密歇根大学，促进科学进步， 许多国家卫生研究院支持的研究人员，以及一批有才华的初级研究人员。

PROJECT SUMMARY/ABSTRACT The University of Michigan (U-M) has established a cutting-edge cryo-electron microscopy (cryo-EM) facility with multiple high-resolution microscopes equipped with the newest direct electron detectors (DEDs). The facility uses automated data acquisition and streamlined image processing protocols, some developed at U-M, which allows for high-throughput structural determination of biological molecules using both single-particle and cryo-electron tomography (cryo-ET) strategies. The combination of improved cameras and automated data collection now makes it routine for users to rapidly collect the data required to determine sub-3.5 Å structures, resolutions that allow for direct model building in the resulting density maps. However, each cryo-EM dataset requiring terabytes of storage and thousands of CPU (Central Processing Unit) or GPU (Graphics Processing Unit) hours on computer systems equipped with large amounts of random-access memory (RAM). This makes high-throughput data collection a new bottleneck in the cryo-EM pipeline. The ability to effectively store, process, and analyze the increasingly massive datasets is essential for this process. Thus, the ability to quickly collect large datasets, combined with a rapidly expanding campus-wide user base, means that access to computational resources now represents a major limiting factor for the success of many U-M cryo-EM projects. The computational resources for cryo-EM at the University of Michigan, housed at the Life Sciences Institute (LSI), were originally funded in part through an S10 grant to the university and will reach their end of life by May 2021 when they will no longer be supported by the vendor. This proposal requests funds to replace the aging computation and storage system, as well as integrate a powerful compute cluster that has nodes with enough memory to efficiently determine atomic resolution structures of large particles (> 1,000 pixel box sizes). This upgraded computation cluster will include 1,248 Intel CPU cores, 16 NVIDIA Quadro RTX 8000 GPUs (Graphic Processing Units), two large memory nodes, and 820 TB of storage, which will be combined and integrated with some of our newer CPU/GPU compute nodes that will remain online. Installation of this resource is essential for: 1) the U-M cryo-EM facility ability to ramp up data collection on four high-resolution cryo-electron microscopes with “on-the-fly” processing; 2) image processing of large heterogeneous cryo-EM datasets (+1,000,000 particles); and 3) compute nodes with enough memory to accommodate the structural analysis of large macromolecular complexes, such as viruses and bacterial secretion systems, and the large volumes used for in situ cryo-ET analysis. The computational infrastructure proposed here will significantly enhance state-of-the-art cryo-EM structure determination at U-M, facilitating the scientific progress of numerous NIH supported investigators, as well as a cohort of talented junior investigators.