Accelerating connectomic proofreading for larger brains and multiple individuals

加速更大大脑和多个个体的连接组学校对

• 批准号: 10413515
• 负责人: Mala Murthy
• 金额: $ 214.39万
• 依托单位: PRINCETON UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-04-01 至 2025-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10413515
• 关键词: 3-Dimensional; Algorithms; Animal Model; Ants; Automation; Bees; Biological; Brain; Cell model; Collaborations; Communities; Computer software; Conflict (Psychology); Data; Data Set; Detection; Development; Drosophila genus; Electrons; Functional disorder; Human; Individual; Infrastructure; Institutes; Institution; International; Internet; Judgment; Manuals; Microscopic; Morphology; Mus; Mutation; Nervous system structure; Neurons; Noise; Readability; Resolution; Scientist; Skeleton; Software Engineering; Speed; Stereotyping; System; Technology; Translating; United States National Institutes of Health; Update; Visualization; Writing; automated segmentation; base; cell type; computer science; connectome; data structure; design; experience; fly; frontier; improved; innovation; large datasets; medical schools; microscopic imaging; neural circuit; novel; open source; prevent; reconstruction; software development; stereotypy; tool

Technology for automating the segmentation of neurons from electron microscopic (EM) data has improved dramatically, making it now possible to obtain accurate reconstructions of neural circuits from large EM volumes. However, even the best automation still must be followed by human proofreading to attain high accuracy. Hundreds of neuroscientists are already using our ChunkedGraph system for proofreading neural circuits. The system is fully web- and cloud-ready, facilitating seamless collaboration. The software is open-source, and at this writing is being operated by three institutions (Princeton, Harvard Medical School, Allen Institute) to serve the proofreading of four large datasets (fly and mouse) by international communities. A notable example is the FlyWire community, which at this writing is engaging over 160 scientists from 40 labs to proofread a whole Drosophila brain. The ChunkedGraph is on its way to becoming a standard and indispensable tool for connectomics. The data structure was designed to permit scaling to arbitrarily large datasets in principle, even to the whole mouse brain connectome project that is currently being considered by the NIH. In practice, there are deficiencies in the current implementation that impede efficiency of proofreading of datasets on the scale of the fly brain, and are preventing further scaling to larger volumes and brains. To remove these barriers to scaling, we will make it possible to upgrade a ChunkedGraph system after proofreading has already started, to take advantage of new and improved automated reconstructions made possible by advances in AI. We will make it possible to visualize neurons in 3D with multi-resolution sharded meshes and skeletons that are rapidly updated after every proofreading edit. We further propose to build a subsequent processing step that rapidly derives morphological features and skeletons, an important prerequisite for downstream analysis and scientific discovery. One of the next frontiers in connectomics is the reconstruction of multiple brains of the same species. For nervous systems with sufficient stereotypy, comparing reconstructions of different individuals can guide the detection and correction of errors. We will develop software that speeds up proofreading by automatically matching a reconstructed neuron to a reference reconstruction, and computing and suggesting candidate corrections if necessary. We will pilot this software for the Drosophila brain, for which multiple EM datasets are now appearing. The same software will be extendable to other model organisms with relatively stereotyped nervous systems (e.g. bee, ant, etc.). In the long term, the approach could further be extended to mammalian nervous systems once the field has developed sufficiently accurate morphological models of cell types. Our proofreading software will continue to be open source and freely accessible. Proofreading generates accurate wiring diagrams of neural circuits, which are helpful for understanding brain function and dysfunction.

通过电子显微镜（EM）数据自动化神经元的分割的技术已改进 急剧上，现在可以从大型EM获得准确的神经回路重建 卷。但是，即使是最好的自动化，仍然必须遵循人类校对才能获得高 准确性。数百名神经科学家已经在使用我们的块状系统来校对神经 电路。该系统已完全可以进行网络和云，可促进无缝协作。该软件是 开源，在撰写本文时，正在由三个机构（普林斯顿，哈佛医学院， 艾伦学院（Allen Institute））由国际社区提供四个大型数据集（Fly and Mouse）的校对。一个 Flywire社区是著名的例子，在撰写本文中，它吸引了160多位科学家，从40个实验室到 校对整个果蝇的大脑。 Chunkedgraph正在成为连接组学的标准和必不可少的工具。数据 结构旨在允许缩放到任意大型数据集，即使是整个鼠标 NIH目前正在考虑的Brain Connectome项目。实际上，有缺陷 当前的实施妨碍数据集校对效率的效率， 正在防止进一步扩展到更大的体积和大脑。要消除这些扩展的障碍，我们将做到这一点 可以在校对启动后升级块状系统，以利用新的优势 并改善了AI的进步使自动重建成为可能。我们将成为可能 用多分辨率的碎片网格和骨骼在3D中可视化神经元，然后迅速更新 每个校对编辑。我们进一步建议建立一个随后的处理步骤，以迅速得出 形态学特征和骨骼，这是下游分析和科学的重要先决条件 发现。 连接组学的下一个前沿之一是重建同一物种的多个大脑。为了 具有足够刻板印象的神经系统，比较不同个体的重建可以指导 检测和纠正错误。我们将开发通过自动加快校对的软件 将重建的神经元与参考重建以及计算和建议候选者匹配 如有必要，校正。我们将为果蝇大脑试用此软件，为此，多个EM数据集是 现在出现。同一软件将可扩展到其他模型生物体，并具有相对刻板印象 神经系统（例如蜜蜂，蚂蚁等）。从长远来看，该方法可以进一步扩展到哺乳动物 神经系统一旦该领域发展了足够准确的细胞类型形态模型。 我们的校对软件将继续是开源并可以自由访问。校对生成 神经回路的准确接线图，有助于理解大脑功能和功能障碍。