冷冻电镜数据采集和实时电子显微图像处理技术研究

A series of breakthrough in electron cryo-microscopy (CryoEM) has significantly improved the resolution of biological structure determination to the near-atomic level. CryoEM is becoming a popular tool in structural biology. However, some bottleneck in CryoEM is limiting its application and developments to the higher level, including the low efficiency in data collection and image processing, which usually takes several weeks or even months. And the new generation of high-speed camera produces huge amount of data, more than 10TB every day, which is the largest challenge to the current system for CryoEM data collection and image processing. The goal of this project will be set to solve and/or improve these problems by developing a new system based on the requirement in acquiring and processing huge dataset, in order to make the entire CryoEM procedure more efficient. Within this system, we are planning to realize the real-time image processing by introducing general-purpose graphic processing unit (GPGPU) into the CryoEM system. The major efforts will be focused on developing new algorithm and approach to implement the CryoEM image processing by GPGPU. Our work will make the near-atomic resolution in CryoEM more obtainable and higher efficient. Meantime, this system will provide a platform for our methodological and technological development in high-resolution CryoEM, and enhance the technical level of Chinese CryoEM society.

冷冻电子显微学借助于最近几年的一系列突破性进展，分辨率获得了极大的提高，其中的一些技术手段已经可以达到近原子分辨率的水平，成为生物结构解析的重要工具。然而，冷冻电镜的数据采集和分析处理仍然是一个费时费力的工作，极大地限制着该项技术的发展和应用。同时，高分辨率所依赖的新一代高速相机，更是产生了巨大的数据量，达到每天10TB以上，这更对现有的数据采集分析系统构成了极大的挑战。为了解决这一问题，本项目将通过发展新一代的冷冻电镜自动化图像采集和实时图像处理系统，来大幅提高整个冷冻电镜应用的效率和易用性。实现基于图形处理器的高速实时图像处理算法将是本项目的研究重点之一。这些算法和软件系统的实现将极大地提高我们现有冷冻电镜系统的使用效率和易用性，使近原子分辨率的结构解析变得更加易于获得。同时这也将为我们在电镜方法和技术研究上提供一个自主研发的平台，提升我国在相应领域的技术开发能力。

在项目的资助下我们开发了一整套冷冻电镜数据采集和分析系统，包括从数据采集到最终的三维重构。在开发的过程中我们引入了新算法思路，来支持自动化的处理分析，包括首次引入和实现基于深度学习的蛋白颗粒识别挑选，基于粒子滤波的高误差容忍度的三维重构参数搜索算法。除了基于图像的结构分析技术，我们针对微晶电子衍射技术开发了数据采集和样品制备技术。通过引入一套适用于普通电镜的样品台相机同步控制机制，大幅降低了冷冻电镜对微晶电子衍射技术的硬件门槛，大幅降低了该技术的应用成本。针对微晶电子衍射样品制备的问题，我们提出和优化了基于聚焦离子束切割的冷冻电镜晶体样品制备方法，消除了微晶电子衍射样品制备的瓶颈问题。目前基于我们开发的微晶电子衍射系统，我们已经解析了多个超高分辨率的结构，而且我们解析的晶体结构质量多优于同期发表的文章。通过整合上述单颗粒和微晶电子衍射技术，本项目实质性地推动了相关技术的发展，使得未来原子分辨率冷冻电镜技术的大规模工业化应用成为可能。