CRCNS: Data Sharing: The EM Open Connectome Project

更广泛的影响：该项目为神经科学界开发开源软件和公共可访问的基础设施，以收集，策划和分析数据密集型集群上的电子显微镜（EM）连接体。公共数据密集型集群，例如我们的Open Connectome Project，减轻了收集数据的实验生物学家的存储管理负担。高通量成像已经产生了大量的数据集，这些数据集压倒了他们实验室的基础设施和专业知识。公共集群还促进了二级数据研究的数据共享，现有结果的验证和重新分析，以及整合和区分从不同受试者，研究人员和仪器收集的多个连接体的多级模型。数据密集型存储和分析将改变EM连接体成像的科学过程。目前，生命科学领域的实验生物学家使用专有的分析工具收集和分析个人的私人数据集。在开放科学方法中，EM连接体数据也远程存储在专为管理和分析大量EM连接体数据而设计的数据密集型计算集群上。开源软件管道自动构建数据产品，包括空间数据库、注释、图形和图形统计。研究人员探索多个连接体。创新的分析技术作为开源软件回馈给社区。在EM Open Connectome中，我们定义了一个框架，让生命科学家、计算机科学家和统计学家组成的跨学科社区参与进来， 解决EM连接体中的两个基本问题：（1）图像分割、注释和跟踪以及（2）图形分析。我们的方法开发了alg-sourcing（算法外包）的概念，研究人员可以轻松地部署，运行，评估和可视化算法对连接体数据库的效率和准确性。EM Open Connectome提供了对数据集和执行框架的访问，因此研究人员只需为其中一个算法任务上传脚本或程序。然后，他们可以获得即时反馈，并可以在数据密集型集群上远程可视化和分析结果，例如，从一个咖啡馆的笔记本电脑上智力优势：主要项目目标是转换从图像数据中提取解剖结构的过程。目前，这是一个手动过程，很少有研究人员探索数十个神经元[6]。EM Open Connectome将支持对收集的最大数据集进行高通量机器注释。障碍包括计算视觉算法的准确性和性能，图像数据的质量以及执行这些分析的软件的访问。我们将探索基于具有解剖先验的多尺度聚集的计算视觉。我们将开发图像处理技术，在计算视觉之前提高数据质量。我们还将构建一个系统工程框架来运行视觉算法，从而实现快速部署、测试和评估。该项目还将通过数据密集型分析加强对大脑功能和计算能力的认识和理解。鉴于空间注册的机器注释，该团队将构建脑图的统计模型，以提供对神经计算的洞察。所有的工具和数据产品都可以向开放科学社区的研究人员公开访问，以便通过合作和跨学科的科学家参与来加速发现。教育和推广：我们的教育使命是促进K-12课程中的数据分析，与国家数学和科学基准保持一致。我们将使用EM Open Connectome提供在线课程计划和活动，直接支持教师需要教授的材料。我们还将开发资源的中心有才华的青年大学预科暑期课程。以博物馆展品和国家科学博览会展位的形式进行的宣传活动支持我们的教育材料和公共数据集。

DESCRIPTION (provided by applicant): Broader Impacts: The project develops open-source software and publicly-accessible infrastructure for the neuroscience community to collect, curate, and analyze electron microscopy (EM) connectomes on data-intensive clusters. Public data-intensive clusters, such as our Open Connectome Project, ease the storage management burden for the experimental biologists that collect data. High-throughput imaging is already producing massive data sets that overwhelm the infrastructure and expertise of their labs. Public clusters also facilitate data sharing for secondary data studies, verification and reanalysis of existing results, and multilevel models that integrate and differentiate multiple connectomes collected from different subjects, researchers, and instruments. Data-intensive storage and analysis will transform the scientific process for EM connectome imaging. At present, experimental biologists in the life sciences collect and analyze individual, private data sets usin proprietary analysis tools. In an Open Science approach, EM connectome data are also stored remotely on a data-intensive compute cluster designed specifically for the curation and analysis of massive EM connectome data. An open-source software pipeline automatically builds data products, including spatial databases, annotations, graphs, and graph statistics. Researchers explore multiple connectomes. Innovative analysis techniques are contributed back to the community as open-source software. In the EM Open Connectome, we define frameworks to engage an interdisciplinary community of life scientists, computer scientists, and statisticians in solving two fundamental problems in EM connectomes: (1) image segmentation, annotation, and tracking and (2) graph analysis. Our approach develops the concept of alg-sourcing (algorithmic outsourcing) in which researchers can easily deploy, run, evaluate, and visualize the efficiency and accuracy of algorithms against connectome databases. The EM Open Connectome provides access to data sets and an execution framework so that researchers simply upload a script or program for one of the algorithmic tasks. Then, they get instant feedback and can visualize and analyze results remotely on the data-intensive cluster, e.g., from a laptop in a cafe. Intellectual Merit: The primary project goal is to transform the process of extracting anatomical structure from image data. Currently, this is a manual process in which few researchers explore tens of neurons [6]. The EM Open Connectome will support high-throughput, machine annotation over the largest data sets being collected. Obstacles include the accuracy and performance of computational vision algorithms, the quality of the image data, and access to software that execute these analyses. We will explore computational vision based on multi-scale aggregates with anatomical priors. We will develop image processing techniques that improve data quality prior to computational vision. We will also build a systems engineering framework to run vision algorithms that allows for rapid deployment, testing, and evaluation. The project will also enhance knowledge and understanding of the functional and computational capabilities of the brain through data-intensive analysis. Given the spatially registered machine annotations, the team will construct statistical models for brain-graphs that provide insight into neural computation. All tools and data products are publicly accessible to an Open-Science community of researchers in order to accelerate discovery through collaboration and by engaging scientists across disciplinary boundaries. Education and Outreach: Our education mission promotes data-analysis in the K-12 curriculum consistent with national benchmarks for math and sciences. We will provide online lesson plans and activities using the EM Open Connectome that directly support the materials that teachers are required to teach. We will also develop resources for the Center for Talented Youth pre-collegiate summer program. Outreach in the form of museum exhibits and a booth at the National Science Fair support our education materials and public data sets.