Motion Sequencing for All: Pipelining, Distribution and Training to Enable Broad Adoption of a Next-Generation Platform for Behavioral and Neurobehavioral Analysis

全民运动测序：流水线、分发和培训，以实现下一代行为和神经行为分析平台的广泛采用

• 批准号: 10616517
• 负责人: Sandeep R Datta
• 金额: $ 46.67万
• 依托单位: HARVARD MEDICAL SCHOOL
• 依托单位国家: 美国
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-04-01 至 2025-01-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10616517
• 关键词: 3-Dimensional; Acceleration; Adoption; Animal Behavior; Animals; Behavior; Behavior assessment; Behavioral; Brain; Calcium; Cloud Computing; Code; Collaborations; Collection; Communities; Complex; Computer software; Corpus striatum structure; Coupling; Data; Data Storage and Retrieval; Dedications; Democracy; Educational workshop; Elements; Gene Expression; Genes; Goals; Head; Imaging technology; Individual; Instruction; Knowledge; Learning; Mathematics; Measures; Methods; Modeling; Modernization; Motion; Mus; Nervous System; Nervous System Physiology; Neurobiology; Neurosciences; Output; Pattern; Persons; Pharmaceutical Preparations; Population; Posture; Procedures; Proliferating; Pythons; Reporting; Resolution; Rodent; Stereotyping; Structure; System; Techniques; Technology; Technology Assessment; Training; Validation; Work; Writing; awake; behavior measurement; cell type; computer science; data format; design; experience; experimental study; flexibility; graduate student; graphical user interface; improved; inhibitor; innovation; insight; machine learning algorithm; machine vision; natural language; neural; neural circuit; neurobehavioral; neuronal circuitry; next generation; open source; sample fixation; skills; tool; unsupervised learning; voltage; web services

Understanding the function of the nervous system requires a sophisticated understanding of its main output, behavior. Although our ability to record from and to manipulate neurons and neural circuits has accelerated at a spectacular pace over the last decade, progress has lagged in coupling the interrogation of the nervous system to similarly high-resolution measures of behavior. As a consequence, we lack a sophisticated understanding of how the brain composes, modifies and controls action. We have recently developed a transformative behavioral characterization technology called Motion Sequencing (MoSeq), which circumvents many of the limitations imposed by typical approaches to behavioral measurement (e.g., overtraining, head-fixation, limited behavioral flexibility). This analytical system works by capturing comprehensive and continuous morphometric data about the three-dimensional (3D) posture of a mouse as it freely behaves. The 3D data are then analyzed using an unsupervised machine learning algorithm to identify patterns of motion that correspond to stereotyped and reused modules of sub-second behavior (which by analogy to natural language we refer to as behavioral “syllables”). The output of this fitting procedure is a parts list for behavior: a limited set of syllables out of which the rodent creates all of its observable action. In addition, within any given experiment MoSeq identifies the specific transition structure (or “grammar”) that places individual syllables into sequences; these sequences encode all patterns of spontaneous behavior expressed by an animal in a given experimental context. We have recently combined this behavioral assessment technology with techniques for neural recording, allowing us to assess the relationship between neural activity in behaviorally-relevant circuits and patterns of action. This combined approach allowed us, for example, to identify a code for elemental 3D pose dynamics in striatum; importantly, these observed correlations validate MoSeq as a technology that enables accurate inference of internal states from external states. However, the code that underlies MoSeq is essentially bespoke, inappropriate for distribution, and difficult for all but expert users to navigate. In addition, implementing MoSeq in its current form requires extensive prior mathematical and computational experience, limiting its use to a small set of users with specialized skills. Here we propose Aims to democratize MoSeq by (1) transforming it into an end-to-end pipeline that can be easily used by graduate-student level neuroscientists with minimal expertise, and which can be modified on an ongoing basis to accommodate improvements to MoSeq and (2) to offer hands-on training in the set-up and appropriate use of MoSeq for characterizing behavior and neural-behavioral relationships. Together these aims will create a vibrant community of MoSeq users; the creation of such a group has the potential to transform the way behavior is analyzed across neuroscience, and promises to lead to broad insights into the many and varied relationships between neural circuits and behavior.

了解神经系统的功能需要对其主要功能有深入的了解 输出、行为。尽管我们记录和操纵神经元和神经回路的能力已经 过去十年以惊人的速度加速，但在将审讯与 神经系统对行为进行类似的高分辨率测量。结果，我们缺乏一个 对大脑如何组成、修改和控制行为的深入理解。 我们最近开发了一种称为 Motion 的变革性行为表征技术 测序 (MoSeq)，它规避了典型行为方法所施加的许多限制 测量（例如过度训练、头部固定、行为灵活性有限）。该分析系统的工作原理是 捕获有关三维 (3D) 姿势的全面且连续的形态测量数据 鼠标自由活动。然后使用无监督机器学习算法分析 3D 数据 识别与亚秒行为的定型和重复使用模块相对应的运动模式 （通过类比自然语言，我们将其称为行为“音节”）。该拟合过程的输出 是行为的组成部分列表：一组有限的音节，啮齿动物用这些音节创造出所有可观察到的动作。 此外，在任何给定的实验中，MoSeq 都会识别特定的转换结构（或“语法”）， 将单个音节放入序列中；这些序列编码所有自发行为模式 由动物在给定的实验环境中表达。我们最近将这种行为结合起来 评估技术与神经记录技术，使我们能够评估之间的关系 行为相关回路和行动模式中的神经活动。这种组合方法使我们能够 例如，识别纹状体中基本 3D 姿态动力学的代码；重要的是，这些观察到的 相关性验证了 MoSeq 是一种能够从外部准确推断内部状态的技术 州。然而，MoSeq 背后的代码本质上是定制的，不适合分发，并且 除了专家用户之外，其他人都很难导航。此外，以当前形式实施 MoSeq 需要 广泛的数学和计算经验，限制其使用到一小部分用户 专业技能。在这里，我们提出通过 (1) 将 MoSeq 转化为端到端的模型来实现 MoSeq 的民主化 具有最少专业知识的研究生水平神经科学家可以轻松使用的管道，并且 可以不断修改以适应 MoSeq 的改进，并且 (2) 提供实践操作 设置和适当使用 MoSeq 来表征行为和神经行为的培训 关系。这些目标共同将创建一个充满活力的 MoSeq 用户社区；建立这样一个 该小组有潜力改变整个神经科学的行为分析方式，并有望引领 广泛洞察神经回路和行为之间的多种关系。