Evaluation of costs and benefits of actions in the basal ganglia

评估基底神经节行动的成本和效益

• 批准号: BB/S006338/1
• 负责人: Rafal Bogacz
• 金额: $ 87.81万
• 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: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=BB%2FS006338%2F1
• 关键词: Evaluation; costs; benefits; actions; basal

We propose to investigate how the brain evaluates the costs and benefits of available options during decision making. We focus on a part of the brain called the basal ganglia that plays an important role in action selection, or a decision about which movement to take. This part of the brain is also involved in the evaluation of options during more abstract decisions. In the basal ganglia there are two main groups of neurons: One group has been shown to facilitate the choice or movement, and we refer to them as the Go neurons, while the other group has been shown to block or prevent movements, and we refer to them as the No-Go neurons. The activity of these two types of neurons is modulated by another population of neurons which release a chemical substance called dopamine. The presence of dopamine increases the activity of Go neurons and decreases the activity of No-Go neurons.Despite years of studies, it is still highly debated what the functions of these two groups of neurons are. One dominant hypothesis is that the while Go neurons facilitate an action, the No-Go neurons block alternative choices, to ensure that only one action is chosen at the time. Alternatively, it has been recently proposed that the Go and No-Go neurons encode the payoffs and costs of actions, while the dopaminergic neurons encode the current motivational state, e.g. hunger. This theory suggests that the basal ganglia circuit weights the payoffs and costs differently according to the motivational state. For example, when an animal is hungry, a high level of dopamine increases the activity of Go neurons and decreases the activity of No-Go neurons, so that the payoffs of actions are weighted more than their costs. However these two hypotheses have not been yet directly tested in experimental data. This proposal aims at answering the fundamental questions concerning the Go and No-Go neurons: what information do they represent, how this information is integrated during choice and modulated by the activity of neurons releasing dopamine, and how the Go and No-Go neurons learn.We propose to record the activity of Go and No-Go and neurons releasing dopamine while mice make choices between two levers associated with different amounts of reward and effort required to obtain it. In most past experiments studying the activity of the Go and No-Go neurons, their activity was recorded via electrodes inserted into animals' brains, but as the Go and No-Go neurons are intermixed in the brain, it is difficult to identify which of them generate the electrical activity. Therefore, in our study we will use a special technique allowing to record the activity of just one group of neurons. To record the Go neurons we will use special genetically modified mice, in which the Go neurons emit light whenever they produce activity. In these mice only the Go neurons emit light, so the light uniquely signals the activity of Go (rather than No-Go) neurons. To measure this light, two optic fibres will be inserted to two sides of the brain. Two additional groups of animals will be used that produce light during the activity of No-Go and neurons releasing dopamine respectively.We will analyse the activity of Go and No-Go neurons during decision making, and investigate if and how they separately encode payoffs and costs of the options chosen by the animal. We will also study how the activity of neurons releasing dopamine influences decision making. Furthermore, by inspecting the changes in the activity over the experiment we will investigate how the Go and No-Go neurons learn about payoffs and costs of actions. Answering these questions is important because action selection and learning in the basal ganglia are affected in Parkinson's disease and several other disorders. Thus understanding how this system operates in the healthy brain is crucial for development of effective treatments that aim at restoring normal function.

我们建议研究大脑如何在决策过程中评估可用选项的成本和收益。我们关注的是大脑中被称为基底节的一部分，它在行动选择或决定采取何种行动方面发挥着重要作用。在做出更抽象的决定时，大脑的这一部分也参与对选项的评估。在基底节中有两个主要的神经元群：一群被证明是促进选择或运动的，我们称之为GO神经元，而另一群被证明是阻断或阻止运动的，我们称之为NO-GO神经元。这两种类型的神经元的活动受到另一组神经元的调节，这些神经元释放一种名为多巴胺的化学物质。多巴胺的存在增加了GO神经元的活性，降低了NO-GO神经元的活性，尽管多年来的研究表明，这两类神经元的功能仍然存在很大争议。一个占主导地位的假说是，虽然Go神经元促进了一个动作，但No-Go神经元阻止了其他选择，以确保当时只选择了一个动作。或者，最近有人提出，Go和No-Go神经元编码行动的回报和成本，而多巴胺能神经元编码当前的动机状态，例如饥饿。这一理论认为，基底节回路根据动机状态的不同，对收益和成本进行不同的权衡。例如，当一只动物饥饿时，高水平的多巴胺会增加GO神经元的活性，降低NO-GO神经元的活性，因此行动的回报比它们的成本更重要。然而，这两个假说还没有在实验数据中得到直接验证。这个建议旨在回答关于GO和NO-GO神经元的基本问题：它们代表什么信息，这些信息在选择过程中是如何整合的，并被释放多巴胺的神经元的活动所调节，以及GO和NO-GO神经元是如何学习的。我们建议记录GO和NO-GO以及释放多巴胺的神经元的活动，同时小鼠在与不同数量的奖励和获得它所需的努力相关的两个杠杆之间进行选择。在过去研究Go和No-Go神经元活动的大多数实验中，它们的活动是通过插入动物大脑的电极来记录的，但由于Go和No-Go神经元在大脑中混合在一起，很难识别它们中的哪一个产生了电活动。因此，在我们的研究中，我们将使用一种特殊的技术，只允许记录一组神经元的活动。为了记录GO神经元，我们将使用特殊的转基因小鼠，在这些小鼠中，GO神经元在产生活动时会发光。在这些小鼠中，只有GO神经元发光，所以光是GO(而不是NO-GO)神经元活动的独特信号。为了测量这种光，将在大脑的两侧插入两根光纤。另外两组动物将被使用，它们分别在No-Go活动和神经元释放多巴胺的活动中发光。我们将分析Go和No-Go神经元在决策过程中的活动，并调查它们是否以及如何分别编码动物选择的选项的收益和成本。我们还将研究释放多巴胺的神经元的活动如何影响决策。此外，通过检查实验中活动的变化，我们将调查Go和No-Go神经元如何了解行动的回报和成本。回答这些问题很重要，因为在帕金森氏病和其他几种疾病中，基底节的动作选择和学习都会受到影响。因此，了解这个系统是如何在健康的大脑中运行的，对于开发旨在恢复正常功能的有效治疗方法至关重要。

项目成果

Preventing Deterioration of Classification Accuracy in Predictive Coding Networks

• DOI: 10.48550/arxiv.2208.07114
• 发表时间: 2022-08
• 作者: Paul F Kinghorn;Beren Millidge;C. Buckley

Designing Ecosystems of Intelligence from First Principles

• DOI: 10.1177/26339137231222481
• 发表时间: 2022-12
• 期刊: Collective Intelligence
• 作者: Karl J. Friston;M. Ramstead;Alex B. Kiefer;Alexander Tschantz;C. Buckley;Mahault Albarracin;R. J. Pitliya;Conor Heins;Brennan Klein;Beren Millidge;D. A. R. Sakthivadivel;T. S. C. Smithe;Magnus T. Koudahl;Safae Essafi Tremblay;C.G. Petersen;K. Fung;Jason G. Fox;S. Swanson;D. Mapes;Gabriel Ren'e

Dynamic control of decision and movement speed in the human basal ganglia.

• DOI: 10.1038/s41467-022-35121-8
• 发表时间: 2022-12-07
• 期刊: NATURE COMMUNICATIONS
• 影响因子: 16.6
• 作者: Herz, Damian M.;Bange, Manuel;Gonzalez-Escamilla, Gabriel;Auer, Miriam;Ashkan, Keyoumars;Fischer, Petra;Tan, Huiling;Bogacz, Rafal;Muthuraman, Muthuraman;Groppa, Sergiu;Brown, Peter
• 通讯作者: Brown, Peter

Robust Graph Representation Learning via Predictive Coding

• DOI: 10.48550/arxiv.2212.04656
• 发表时间: 2022-12
• 期刊: ArXiv
• 作者: Billy Byiringiro;Tommaso Salvatori;Thomas Lukasiewicz