Towards a 'Big Data Browser' for standardised datasets in neuroscience, and its application in ion channel and single cell modelling.

面向神经科学标准化数据集的“大数据浏览器”及其在离子通道和单细胞建模中的应用。

• 批准号: BB/N019512/1
• 负责人: Tim Philipp Vogels
• 金额: $ 85.84万
• 依托单位: University of Oxford
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2017
• 资助国家: 英国
• 起止时间: 2017 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=BB%2FN019512%2F1
• 关键词: Towards; Big; Data; Browser; standardised

Big data, and with it complex, multifaceted, publicly available data sets, has arrived in modern neuroscience research over the last decade. Unlike other fields however, the datasets in neuroscience are very heterogeneous and often require heavy post-processing and contextual annotation. This enrichment of metadata presents a challenge in itself because any data needs to be filtered and visualised with changing situational requirements, and their high dimensionality often make these data sets unruly and difficult for traditional static visualisation techniques. In preliminary work (submitted), we have manually annotated, enriched and standardised a dataset of over 2000 ion channel models according to their published context and their stereotypical response to predefined test protocol. We then developed a ``big data browser'' to allow users to quickly browse all ion channel models and visualise their functional characteristics with interactive filtering. These ion channel models have become the standard in the field for creating experimentally-constrained neuron and network models. Our method provides all necessary information for users searching for specific ion channel types to narrow down the circle of candidate models and can save an individual researcher as much as 4 weeks of preliminary literature research and test simulations.In this proposal, we aim to expand our work. Divided into three work packages, we aim to maintain and expand this resource in the context of ion channel models, as well as to make it a general case browser for neuroscience data. In WPI, we plan to expand the database to be simulator-independent and to include experimental data, making it more comprehensive and allow direct comparison of experiments to models. This will make the resource a useful tool for theorists as well as experimentalists. Next, we plan to integrate our database with other neuroscience resources, and maintain this all in one comprehensive and up-to-date platform. In this way, we can take advantage of existing frameworks and act as a complement to them.In WPII, we aim to expand the current functionality of the database to include multi-channel dynamics, multi-compartmental models, and finally to combine these two together to allow for analysis of complex dynamics. We will use the existing framework developed in our pilot work and WPI. Our ultimate goal is to produce a tool for the complete creation and testing of neuron models with complex morphology and channel composition. This will be of great benefit to experimentally-constrained modelling, making the creation of models easier and more intuitive. Testing a hypothesis of functional deficits based on proposed ion channel disfunction or morphological aberration will only take a matter of hours instead of several weeks.In WPIII, we aim to expand our data browser for ion channel models to other areas in neuroscience by making it a general 'Big Data Browser'. The main issue with big data currently is to create graphical interfaces that allow the exploration of complex datasets without requiring expert knowledge. We hope to create a unified interface for the storage, organisation and visualisation of datasets such as these, in order to make data more accessible and reusable. The end result will be an out-of-the-box data visualiser that can be adapted to any new dataset, and even to allow the comparison between datasets in an easy way.In summary, our proposal will provide the field with a resource for building detailed neuron models with complex ion channel dynamics and comparing them directly to experiments. In a second step, it will also provide the field with a much-needed data interface, adaptable to any dataset, solving a field-wide problem of how to efficiently share and compare heavily annotated big data sets.

在过去的十年里，大数据以及复杂的、多方面的、公开可用的数据集已经进入了现代神经科学研究领域。然而，与其他领域不同的是，神经科学中的数据集是非常不同的，通常需要大量的后处理和上下文注释。这种丰富的元数据本身就是一个挑战，因为任何数据都需要随着不断变化的情况要求进行过滤和可视化，而且它们的高维度经常使这些数据集变得难以控制，难以使用传统的静态可视化技术。在前期工作(提交)中，我们根据公布的背景和对预定义测试协议的刻板反应，对2000多个离子通道模型的数据集进行了手动注释、丰富和标准化。然后，我们开发了一个“大数据浏览器”，允许用户快速浏览所有离子通道模型，并通过交互过滤可视化它们的功能特征。这些离子通道模型已经成为该领域的标准，用于创建实验受限的神经元和网络模型。我们的方法为搜索特定离子通道类型的用户提供了所有必要的信息，以缩小候选模型的范围，并可以为单个研究人员节省多达4周的前期文献研究和测试模拟。分为三个工作包，我们的目标是在离子通道模型的背景下维护和扩展这一资源，并使其成为神经科学数据的通用案例浏览器。在WPI中，我们计划将数据库扩展为独立于模拟器并包括实验数据，使其更加全面，并允许直接将实验与模型进行比较。这将使该资源成为理论家和实验者的有用工具。下一步，我们计划将我们的数据库与其他神经科学资源整合，并在一个全面和最新的平台上维护所有这些资源。在WPII中，我们的目标是扩展数据库的当前功能，以包括多通道动力学、多间隔模型，并最终将这两者结合在一起，以允许分析复杂的动力学。我们将使用在我们的试点工作中开发的现有框架和WPI。我们的最终目标是制造一个工具，用于完整地创建和测试具有复杂形态和通道组成的神经元模型。这将对实验约束的建模有很大的好处，使模型的创建更容易和更直观。根据提出的离子通道功能障碍或形态异常来测试功能缺陷的假说只需要几个小时，而不是几周。在WPIII中，我们的目标是将我们的离子通道模型数据浏览器扩展到神经科学的其他领域，使其成为一个通用的“大数据浏览器”。目前大数据的主要问题是创建图形界面，允许在不需要专业知识的情况下探索复杂的数据集。我们希望为这些数据集的存储、组织和可视化创建一个统一的接口，以使数据更容易访问和重复使用。最终的结果将是一个开箱即用的数据可视化工具，它可以适应任何新的数据集，甚至允许以一种简单的方式在数据集之间进行比较。总之，我们的提议将为该领域提供一个资源，用于建立具有复杂离子通道动力学的详细神经元模型，并将它们直接与实验进行比较。在第二步，它还将为该领域提供一个急需的数据接口，适用于任何数据集，解决如何有效地共享和比较大量注释的大数据集的整个领域的问题。

项目成果

A meta-learning approach to (re)discover plasticity rules that carve a desired function into a neural network

• DOI: 10.1101/2020.10.24.353409
• 发表时间: 2020-10
• 期刊: bioRxiv
• 作者: Basile Confavreux;Everton J. Agnes;Friedemann Zenke;T. Lillicrap;T. Vogels

Training deep neural density estimators to identify mechanistic models of neural dynamics.

• DOI: 10.7554/elife.56261
• 发表时间: 2020-09-17
• 期刊: eLife
• 影响因子: 7.7
• 作者: Gonçalves PJ;Lueckmann JM;Deistler M;Nonnenmacher M;Öcal K;Bassetto G;Chintaluri C;Podlaski WF;Haddad SA;Vogels TP;Greenberg DS;Macke JH
• 通讯作者: Macke JH