Integrated functional and structural analysis of an entire column in mouse primary visual cortex

项目总结 视皮层中的神经元形成了错综复杂的连接结构和地形排列。这个 众所周知，神经元的结构和形态组织限制了其功能特性。至 了解这些限制，有必要生成大规模的解剖和功能测量 大脑的一部分。电子显微镜(EM)和荧光显微镜正在进行的努力有望大规模地 加快生成此类数据的速度。要发现结构和功能之间的关系，一个 方法是使用大脑中既有结构组件又有功能组件的计算模型。 卷积神经网络(CNN)，以及更广泛的机器学习(ML)已经显示出越来越多的 在对大脑的功能特性进行建模方面有希望。然而，这些基于CNN的模型往往不是 由大脑中神经元的真实结构引导。在这项建议中，我们寻求分享和分析 迄今为止从小鼠初级视觉皮质(V1)的整个柱子收集的最大钙成像数据集。 我们建议将该数据集与单独的公开可用的电子显微镜(EM)数据集整合在一起 使用一种新的基于CNN的拓扑学上可信的神经元模型，在小鼠V1中的体积为1mm3。我们将使用 这一新模型有助于理解结构性组织如何发挥作用。我们的方法是将 对神经网络中的计算进行研究的拓扑建模。钙成像数据集包括 四只小鼠的体积双光子和三光子成像。该数据包含来自 初级视皮层800um×800um×~800um区域内的神经元，横跨所有视层 从Pia到白质。我们建议使用这个数据集来系统地表征生理学 小鼠V1整柱内空间选择性神经元和运动选择性神经元的特性。我们会 确定空间选择性神经元中开和关亚场的组织和大小是否以及如何 显示视觉层之间的差异。然后我们将测试空间选择性和运动的假设- 选择性神经元在V1内形成不同的网络。我们建议将这两个网络合并为 神经元的卷积神经网络模型，以测试该模型是否会导致更高的预测精度 以及对神经模式选择性的更好估计。最后，我们建议将分支主题描述为 使用EM重建小鼠视皮质体积为1mm3的V1。然后我们将整合 这些基序与基于CNN的神经元模型相结合，能够准确预测钙离子成像反应。我们会 使用此设置来测试在模型中合并分支主题是否会产生更准确的预测 神经活动和更好地估计神经模式选择性。我们的提案提供了一种系统的方法 整合结构和功能数据集，以了解结构如何提供视觉皮质的功能。

PROJECT SUMMARY Neurons in the visual cortex form an intricate connectivity structure and topographic arrangement. The structural and morphological organization of the neurons is known to constrain its functional properties. To understand these constraints, it is necessary to generate large-scale anatomical and functional measurements of the brain. Ongoing efforts in electron microscopy (EM) and fluorescent microscopy promise to massively accelerate the speed of generating such data. To discover the relationship between structure and function, one approach is to use models of computation in the brain that have both structural and functional components. Convolutional neural networks (CNNs), and more broadly machine learning (ML), have shown an increasing promise in modeling the functional properties of the brain. However, these CNN-based models are often not guided by the real structure of the neurons in the brain. In this proposal, we seek to share and analyze the largest calcium imaging dataset collected to date from an entire column in mouse primary visual cortex (V1). We propose to integrate this dataset with a separate publicly available electron microscopy (EM) dataset of 1mm3 volume in mouse V1 using a novel topologically-plausible CNN-based model of neurons. We will use this new model to understand how the structural organization affords function. Our approach is to bring topological modeling to the study of the computation in neural networks. The calcium imaging dataset includes volumetric 2-photon and 3-photon imaging in four mice. The data contains recordings of visual responses from neurons within an 800um × 800um × ~800um region of the primary visual cortex, spanning all visual layers from pia to white matter. We propose to use this dataset to systematically characterize the physiological properties of spatially-selective and motion-selective neurons in an entire column in mouse V1. We will determine whether and how the organization and size of ON and OFF subfields in spatially-selective neurons reveal differences across visual layers. We will then test the hypothesis that the spatially-selective and motion- selective neurons form distinct networks within V1. We propose to incorporate these two networks into convolutional neural network model of neurons to test whether this model results in higher predictive accuracy and better estimation of neural pattern selectivity. Finally, we propose to characterize the branching motifs in mouse V1 using the EM reconstructions from 1mm3 volume in mouse visual cortex. We will then integrate these motifs with the CNN-based models of neurons to accurately predict calcium imaging responses. We will use this setup to test whether incorporating the branching motifs in a model result in more accurate prediction of neural activity and better estimation of neural pattern selectivity. Our proposal offers a systematic approach to integrate structural and functional datasets to understand how structure affords function in the visual cortex.