Mozak: Creating an Expert Community to accelerate neuronal reconstruction at scale

项目摘要 这个项目的目的是利用计算和人类在神经元重建方面的最佳专业知识。 加速从国际来源的成像数据中发现全球神经科学的目标。我们 建议创建一个基于云的统一平台，用于将神经元的三维图像融合到单个 分析平台，以(1)培训和发展新的全球重建者专家社区，以便在 来自这些小组的数据，以(2)生成开放的社区来源的神经元重建数据库 可以合并到神经元互连的三维地图中的成像数据--在该地图上(3) 可以覆盖新的注释和更复杂的功能和分子数据。我们的方法将会发展 随着时间的推移，神经科学界的需求不断增长。要做到这一点，在目标一号(神经元 大规模重建)，我们将测试新开发的基于众包游戏的平台Mozak是否可以 大规模培养一批新的人类专家，能够加快当前的重建速度 至少提高了一个数量级，同时增加了 最后的重建。在目标二(稳健的多用途注释)中，我们将增强基本神经元 通过添加特定的语义标注进行重建--包括体块和形态 要求量化、体积分析和持续特征(例如树突棘、轴索静脉曲张) 来自神经科学界。经验丰富的高级会员将有机会 通过日益复杂的神经元进入全乔木全脑神经元投射和多个 局部环路中聚集的神经元组。最后，在目标3(创建一项研究--适应性数据 库)，我们的目标是开发一个使用Mozak接口重建的神经元图像的数据库，该数据库 将直接服务于不同研究小组的一般和具体需求。我们的目标是让这一切 数据库动态自适应-因为新的研究问题总是会带来额外的新需求 与其他数据模式的注释和交叉引用。这一高质量、无偏见的处理存储库 也将非常适合用于训练集的自动化算法，并生成三维的 阿伦脑科学研究所(AIBS)等地图的共同坐标框架。我们预计， 随着新的大型语料库的出现，计算重建方法将进一步完善 重建。总的来说，完成这三个目标将创建一个分析套件以及 能够对大规模数据集进行深入分析的在线专家社区 显著加快神经科学研究，增强用于重建分析的机器学习，以及 创建基线神经元形态数据的通用平台，以对照功能异常的神经元 可以分析。

Project Summary This project aims to leverage the best of both computational and human expertise in neuronal reconstruction towards the goal of accelerating global neuroscience discovery from internationally-sourced imaging data. We propose to create a cloud-based unified platform for converging 3-dimensional images of neurons onto a single analysis platform to (1) train and grow a new expert community of global reconstructors to work across the data from these groups, to (2) generate a community-sourced neuronal reconstruction database of open imaging data that can be incorporated into a 3-dimensional map of neuronal interconnectivity - onto which (3) novel annotations and more complex functional and molecular data can be overlaid. Our approach will evolve with the growing needs of the neuroscience community over time. To do this, in Aim One (Neuronal Reconstruction at Scale), we will test if the newly developed crowd-sourced game-based platform Mozak can develop a collective of new human experts at scale, capable of accelerating the rate of current reconstruction by at least an order of magnitude, at the same time as increasing the robustness, quality and unbiasedness of the final reconstructions. In Aim Two (Robust Multi-Purpose Annotation), we will enhance basic neuronal reconstruction by adding specific semantic annotation— including soma volume and morphological quantification, volumetric analysis, and ongoing features (e.g. dendritic spines, axonal varicosities) requested from the neuroscience community. Experienced and high-ranking members will be given the opportunity to advance through increasingly complex neurons into full arbor brain-wide neuronal projections and multiple clustered groups of neurons in localized circuits. Finally, in Aim 3 (Creation of a Research-Adaptive Data Repository), we aim to develop a database of neuronal images reconstructed using the Mozak interface that will directly serve the general and specific needs of different research groups. Our goal is to make this database dynamically adaptive — as new research questions will invariably bring new needs for additional annotations and cross-referencing with other data modalities. This highquality unbiased processing repository will also be perfectly suited for training sets for automated algorithms, and the generation of a 3-dimensional maps such as Allen Institute for Brain Science (AIBS) common coordinate framework. We expect that the computational reconstruction methods will further improve with the new large corpus of “gold standard” reconstructions. Collectively, the completion of these three aims will create an analysis suite as well as an online community of experts capable of performing in depth analysis of large-scale datasets that will significantly accelerate neuroscience research, enhance machine learning for reconstruction analysis, and create a common platform of baseline neuronal morphology data against which aberrantly functioning neurons can be analyzed.