Interaction of external inputs with internal dynamics: influence of brain states on neural computation and behavior

总体--外部输入与内部动态的交互作用： 大脑状态对神经计算和行为的影响 项目摘要 神经科学中的一个中心挑战涉及了解皮质神经元的组装是如何组成的 不同的细胞类型和居住在不同的层，共同工作以产生相干的动态内态， 然后与外部感觉输入相互作用，每时每刻产生状态依赖的行为 基础。迎接这一基础性挑战的主要障碍包括缺乏足够的技术和 用于监视、控制、识别和模拟从皮层细胞产生的神经状态动力学的计算工具 跨越多个皮质细胞类型和层的组件。我们提议发展一种前所未有的融合 通过建立在我们团队先前重大工作的基础上，实现这些能力的技术和计算能力。在……里面 具体地说，我们的技术和计算平台相结合将使我们能够：(1)执行体成像 收集相关的时空活动模式和3D 定位；(2)同时将任意时空图案写入数十到数百个单独的 使用双光子多SLM方法以毫秒的时间分辨率识别细胞；以及(3)使用水凝胶 组织化学和单细胞测序方法一样，可以获得深层分子细胞类型信息 在行为过程中被测量和控制的神经元。史无前例的同步 读/写/单元打字技术将与计算方法紧密集成，这些方法可以：(1)使用 识别和提取神经状态和动力学规律的最新系统辨识方法 管理它们与外部输入的相互作用；以及(2)在可能的天文数字中 时空刺激模式，预测可以最好地改进模型的有趣模式，产生概念 洞察力，并产生最优控制皮质回路动力学的能力，具有潜在的临床意义。 这种技术和计算的结合将支持下一代实验，使我们能够学习 大脑皮层回路的动态语言(就状态空间动力学而言)，回放它的修改版本 洞察和控制的语言，并理解这种语言是如何从 跨层的多种单元格类型。我们的技术/计算平台将在多个实验中得到验证 跨物种和大脑区域，在深入和长期的内部状态动力学理论的指导下 计算神经科学。在整个过程中，新方法将在不同的 我们实验实验室的准备工作(这种横向互动显示在蓝色文本中)。尤其是我们 我将重点测试神经计算的几个基本类别所依据的理论：(1)感觉能力 通过检测和放大自发的微弱感觉输入来产生准确感知的网络 背景活动；(2)多感觉输入的贝叶斯集成，将感觉运动体验转换为 外部状态变量及其不确定性的内部估计；以及(3)触发和维护 能够控制稳定行为的离散内部吸引子状态。

Overall - Interaction of external inputs with internal dynamics: influence of brain states on neural computation and behavior Project Summary A central challenge in neuroscience involves understanding how assemblies of cortical neurons, comprised of different cell types and inhabiting different layers, work together to generate coherent dynamical internal states, that then interact with external sensory inputs to generate state-dependent behaviors on a moment-by-moment basis. Key impediments to meeting this foundational challenge include lack of adequate technological and computational tools to monitor, control, identify and model neural state dynamics emerging from cortical cell assemblies spanning multiple cortical cell-types and layers. We propose to develop an unprecedented confluence of technology and computation to achieve such capabilities by building on our team’s significant prior work. In particular, our combined technology and computation platform will enable us to: (1) perform volumetric imaging of thousands of cortical cells during behavior to collect both relevant spatiotemporal activity patterns and 3D positioning; (2) simultaneously write arbitrary spatiotemporal patterns into tens to hundreds of individually identified cells at millisecond temporal resolution using 2-photon multiSLM methods; and (3) using hydrogel tissue-chemistry and single-cell sequencing methods, obtain deep molecular cell-type information in the same neurons that were both measured and controlled during behavior. This unprecedented simultaneous read/write/cell-typing technology will be tightly integrated with computational methods that can: (1) employ state of the art systems identification methods to identify and extract neural states and the dynamical laws governing their interactions with external inputs; and (2) amongst the astronomical number of possible spatiotemporal stimulation patterns, predict interesting ones that can best refine models, yield conceptual insights, and yield the capacity for optimal control of cortical circuit dynamics, with potential clinical relevance. This combined technology and computation will empower next-generation experiments that allow us to learn the dynamical language (in terms of state space dynamics) of cortical circuits, play back modified versions of this language for both insight and control, and understand how this language emerges from the concerted activity of multiple cell-types across layers. Our technology/computation platform will be validated in multiple experiments across species and brain regions, guided by deep and long-standing theories of internal state dynamics in computational neuroscience. Throughout, new methods will be collaboratively validated in the diverse preparations of our experimental labs (such cross-cutting interactions are shown in blue text). In particular we will focus on testing theories underlying several foundational classes of neural computation: (1) ability of sensory networks to generate accurate percepts by detecting and amplifying weak sensory inputs amidst spontaneous background activity; (2) Bayesian integration of multisensory inputs to convert sensorimotor experiences into internal estimates of external state variables and their uncertainty; and (3) triggering and maintenance of discrete internal attractor states capable of controlling stable behavior.