Psych-DS: A FAIR data standard for behavioral datasets

摘要： 行为数据是生物医学研究的核心，包括两种同步测量(例如大脑激活 以及在fMRI扫描中从阅读任务中按下按钮)，以及独立执行的那些(例如识字 调查问卷。)与神经生理学和脑成像数据相比，行为数据往往相对较小， 实验脚本和结果数据集的文件大小均以MB为单位，而不是TB。这个 行为数据的关键挑战不是大小，而是可变性：许多研究的设计并不公平(可发现， 可访问、可互操作和可重用)共享或自动化分析管道。那些是 机器可读缺乏跨实验室甚至是跨实验室的通用格式、组织或文档标准 个别研究。 我们建议在大脑主动性中实施人类行为实验的标准 社区开发的数据规范，智力-DS。因为Squch-DS被设计成类似于RAW 研究人员已经获得的数据，这项工作将导致广泛的用户池采用该标准，而不是 对现有工作流程的额外工作或中断；这些研究人员的反馈将是 发展提案的所有三个目标： 首先，我们将为Squch-DS数据集创建验证器软件，以确保实现标准的工具 对于使用各种语言(Python、R、Java脚本)的研究人员以及 手工汇编数据集。其次，标准包和验证包将被集成到至少三个 心理学家和认知神经学家广泛使用的流行实验演示包 和Lookit/儿童帮助科学)。第三，支持涉及协调一致的项目 行为和神经成像/神经生理学数据，我们将实现工具在心理障碍之间进行转换 和脑成像数据标准(BIDS)。 在如何组织行为数据方面缺乏共识，阻碍了广泛而多样的 与生物医学相关的研究从与目前正在建设的基础设施完全集成到 支持大规模的神经科学研究。实施和采用Squch-DS标准将提供 一种机器可读的数据格式，可用于各种研究背景，并为 为进一步提高生物医学和行为学的标准化和重现性奠定基础 科学。

Summary: Behavioral data is central to biomedical research, including both synchronous measures (e.g. brain activation and button-presses from a reading task in an fMRI scan), and those performed independently (e.g. a literacy questionnaire.) Compared to neurophysiology and brain imaging data, behavioral data is often relatively small, with file sizes in the megabytes rather than terabytes for both experimental scripts and resulting datasets. The key challenge for behavioral data is not size but variability: many studies are not designed with FAIR (Findable, Accessible, Interoperable, and Reusable) sharing or automated analysis pipelines in mind. Those that are machine-readable lack common standards of format, organization, or documentation across labs or even individual studies. We propose to implement a standard for human behavioral experiments in the BRAIN initiative using a community-developed data specification, Psych-DS. Because Psych-DS is designed to be similar to the raw data researchers already acquire, this work will result in a broad pool of users adopting the standard without additional effort or interruption to existing workflows; feedback from these researchers will be central to development of all three aims of the proposal: First, we will create validator software for Psych-DS datasets, to ensure that tools implementing the standard remain cross-compatible for researchers using a variety of languages (Python, R, Javascript) as well as compiling datasets by hand. Second, the standard and validation packages will be integrated into at least three popular experiment presentation packages in wide use by psychologists and cognitive neuroscientists (jsPsych, PsychoPy, and Lookit/Children Helping Science). Third, to support projects that involve coordinated behavioral and neuroimaging/neurophysiological data, we will implement tools to translate between Psych-DS and the Brain Imaging Data Standard (BIDS). The lack of consensus around how behavioral data are organized prevents a broad and diverse swath of biomedically relevant research from being fully integrated with the infrastructures currently being built to support neuroscientific research at scale. Implementation and adoption of the Psych-DS standard will provide a machine-readable data format that can be used across a wide variety of research contexts, and lay the groundwork for further improvements to standardization and reproducibility in the biomedical and behavioral sciences.