Contextual modulation of visual decision-making across the visual hierarchy

项目摘要/摘要 在感官决策中，选择受到非感官因素的影响，如 动机、注意力和最近的审判历史。我们寻求将这些影响融入到一个 漂移扩散模型(DDM)，通过将非感觉变量建模为确定性的调节器 感官证据积累的起点或漂移率。然而，决策 由于模型的非平稳性和长期相关性，模型容易受到混淆 行为数据。需要更多的工作来量化这些特性并开发新的统计方法 克服这些问题的方法。现有数据集已被证明不够用，因此新的数据集 必须被收集起来。为了更深入地了解语境调节的神经机制 我们的目标是比较非感官影响对感官决策的影响- 在视觉层次的各层之间以及在并行的视觉流之间进行。我们 有证据表明存在差异，这可以被用来确定大脑中的哪个部位 非视觉信息进入视觉决策。为了确定R01的可行性，我们 需要与统计学家建立新的合作关系，制定培训方法 多个视觉任务中的单个动物；表明视觉任务对非视觉任务的敏感性不同 感觉调节；并表明我们可以获得所需的试验数据来匹配和 比较受试者内部的模型。我们建议训练单个动物的格栅方向， 随机点运动、对象标识、亮度或对比度的空间位置以及引导 两个新任务(随机漂移栅格、运动空间定位)。这些视觉功能包括 被认为是在不同的大脑区域提取的：在初级视觉皮质(V1)；在不同的高级 腹侧或背侧视束的视区；或V1非依赖的丘脑通路。我们 将使用自动化的高吞吐量机笼收集交错任务的长期数据 检验，并验证这些数据是否符合模型拟合的统计要求。老鼠是 这是这项研究的理想选择，因为单个大鼠可以学习多种视觉任务，而我们能够 在没有水或食物限制的情况下，每只大鼠获得105-106次行为试验。老鼠也适合于 病毒载体打靶策略与光遗传学高密度电生理学 表现得像动物一样。在这个为期2年的R34项目结束时，实验室将组装一个新的 多学科研究团队准备剖析潜在的电路机制， 经过验证的视觉任务、训练方案、统计方法和模型拟合方法。 这些准备工作将支持Brain Initiative：针对BCP R01的应用程序，旨在 剖析神经表征和语境调节选择的电路机制。

Project Summary/Abstract In sensory decision-making, choices are influenced by non-sensory factors such as motivation, attention, and recent trial history. We seek to incorporate these influences into a drift diffusion model (DDM), by modeling non-sensory variables as deterministic modulators of the starting point or drift rate of sensory evidence accumulation. However, decision-making models are subject to confounds due to the non-stationarity and correlations in long-term behavioral data. More work is needed to quantify these properties and develop new statistical approaches to overcome them. Existing datasets have proven inadequate, so new datasets must be collected. To gain insight into the neural mechanisms of contextual modulation in decision-making, our goal is to compare non-sensory influences on sensory decision- making across levels of the visual hierarchy and between parallel visual streams. We have evidence there are differences, which could be leveraged to identify where in the brain non-visual information enters into visual decision-making. To establish feasibility for an R01, we need to establish a new collaboration with a statistician, develop methods for training single animals in multiple visual tasks; show that visual tasks differ in their sensitivity to non- sensory modulation; and show that we can obtain the amount of trial data required to fit and compare models within subject. We propose to train individual animals in grating orientation, random-dot motion, object identity, spatial location of luminance or contrast, as well as piloting two new tasks (stochastic drifting grating, spatial location of motion). These visual features are thought to be extracted in different brain areas: in primary visual cortex (V1); in different higher visual areas (HVAs) in ventral or dorsal streams; or in a V1-independent collicular pathway. We will collect long-term data on the interleaved tasks using automated high-throughput in-cage testing, and validate that these data meet statistical requirements for model fitting. Rats are ideal for this study because individual rats can learn multiple visual tasks, and we are able to obtain 105-106 behavioral trials per rat without water or food restriction. Rats are also suitable for viral vector targeting strategies and high-density electrophysiology with optogenetics in freely behaving animals. At the end of this 2-year R34 project, the lab will have assembled a new multi-disciplinary research team poised for dissection of underlying circuit mechanisms, with validated visual tasks, training protocols, statistical approaches, and model-fitting methods. These preparations will support a BRAIN Initiative: Targeted BCP R01 application aimed at dissecting neural representations and circuit mechanisms of contextual modulation of choice.