Data-Science Core

数据科学核心

• 批准号: 10615033
• 负责人: Jan Drugowitsch
• 金额: $ 35.98万
• 依托单位: UNIVERSITY OF ROCHESTER
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-08-01 至 2025-04-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10615033
• 关键词: Architecture; Behavioral; Brain; Charge; Code; Collaborations; Communities; Computer Models; Computer software; Data; Data Analyses; Data Science Core; Data Scientist; Data Storage and Retrieval; Development; Documentation; Ensure; Experimental Designs; Factor Analysis; Future; Goals; Information Technology; Infrastructure; International; Investments; Joints; Label; Laboratories; Metadata; Methods; Modeling; Morphologic artifacts; Nature; Neurosciences; Population; Principal Component Analysis; Process; Publishing; Reproducibility; Research; Research Personnel; Resources; Software Tools; Source Code; Standardization; Statistical Models; Stimulus; Structure; Techniques; Testing; Time; Variant; Work; computer infrastructure; data access; data cleaning; data exchange; data infrastructure; data sharing; data standards; experience; experimental study; improved; invention; member; multidisciplinary; neural; neural model; neuromechanism; neurophysiology; open source; repository; searchable database; theories; tool; voltage

Project Summary This application proposes to use theory-driven experimental design, with advanced techniques for neural recording, data analysis, and computational modeling, to investigate the neural mechanisms, circuits, and representations underlying the perceptual process of causal inference in space and time. The multidisciplinary nature of the proposed work requires close collaboration among consortium members. The Data Science Core will facilitate this collaboration and provide the tools necessary to handle and analyze the large-scale neural data collected in the proposed experiments. To achieve these goals, the Data Science Core will rely on existing infrastructure, open standards, and open-source software as much as possible. Aim 1 will establish a unified data standard, and data exchange and storage infrastructure, using the architecture established by the International Brain Laboratory, which stores metadata in a relational, searchable database, and experimental and processed data on a separate file server. Github will enable joint development, exchange, and documentation of the code underlying data preprocessing, processing, and analysis. To relate data to models, voltages recorded experimentally must be transformed into standardized spike times and counts, without artifacts or confounds. Aim 2 will develop a principled, transparent, and reproducible pipeline for this preprocessing and apply it to all neurophysiological data generated in Projects B and C. The first stage will eliminate electrical and behavioral artifacts and convert voltages into spike times and local field potentials. The second stage will use a statistical model of neural activity to identify and label potential outliers. This pipeline will produce annotated and cleaned data in a standardized format that can be used to perform reliable analyses, model fitting, and hypothesis tests. Aim 3 will combine cutting-edge methods and convert them to software tools that can be reliably applied to new data. Most of this effort will be applied to variants of latent- state discovery techniques that jointly fit the influence of stimuli, model-driven hypothesized latent states, and unobserved latent states such as slow fluctuations. The central work of this aim is to implement those tools, help the team apply them to the data generated by the collaboration, and refine them for public use. Aim 4 is to share the experimental data with the wider research community by uploading the relevant portions of the data to public and freely accessible repositories. Code, documentation, and use cases will be made public on Github. The use of standard data structures, open standards, and open-source software will ensure barrier-free access, ease of use, and reproducibility for neuroscience researchers. With the help of a full-time data scientist hired to manage these efforts, the Data Science Core will build on established data storage and analysis standards and methods to produce cleaned and standardized data that our consortium can use to close the loop between theory and experiments. By sharing code, use cases, and data with other researchers, this project will also improve and extend these resources for future use by others.

项目概要 该应用建议使用理论驱动的实验设计，以及先进的神经网络技术 记录、数据分析和计算建模，以研究神经机制、电路和 空间和时间中因果推理的感知过程的表征。多学科 拟议工作的性质需要联盟成员之间的密切合作。数据科学核心 将促进这种合作，并提供处理和分析大规模神经网络所需的工具 在拟议的实验中收集的数据。为了实现这些目标，数据科学核心将依赖于 尽可能利用现有基础设施、开放标准和开源软件。目标 1 将建立一个 统一数据标准、数据交换和存储基础设施，采用由联盟建立的架构 国际脑实验室，将元数据存储在可搜索的关系数据库和实验数据库中 并在单独的文件服务器上处理数据。 Github 将实现联合开发、交流和 数据预处理、处理和分析的代码文档。为了将数据与模型联系起来， 实验记录的电压必须转换为标准化尖峰时间和计数，而无需 伪影或混淆。目标 2 将为此开发一个有原则的、透明的、可重复的管道 预处理并将其应用于项目 B 和 C 中生成的所有神经生理学数据。第一阶段将 消除电气和行为伪影，并将电压转换为尖峰时间和局部场电位。这 第二阶段将使用神经活动的统计模型来识别和标记潜在的异常值。这条管道 将以标准化格式生成带注释和清理的数据，可用于执行可靠的操作 分析、模型拟合和假设检验。目标 3 将结合前沿方法并将其转化为 可以可靠地应用于新数据的软件工具。这项工作的大部分将应用于潜在的变体 状态发现技术共同适应刺激的影响、模型驱动的假设潜在状态，以及 未观察到的潜在状态，例如缓慢波动。这一目标的核心工作是实施这些工具， 帮助团队将它们应用到协作生成的数据中，并对其进行提炼以供公众使用。目标 4 是 通过上传数据的相关部分与更广泛的研究社区共享实验数据 到公共和可免费访问的存储库。代码、文档和用例将在以下位置公开 GitHub。使用标准数据结构、开放标准、开源软件，确保无障碍 神经科学研究人员的访问性、易用性和可重复性。在全职数据科学家的帮助下 受聘来管理这些工作的数据科学核心将建立在已建立的数据存储和分析的基础上 生成清理和标准化数据的标准和方法，我们的联盟可以使用这些数据来关闭 理论与实验之间的循环。通过与其他研究人员共享代码、用例和数据，这 项目还将改进和扩展这些资源以供其他人将来使用。