CRCNS: Neural coding and computation in large ensembles in prefrontal cortex

项目摘要 认知的本质是选择，要理解选择，我们需要理解在复杂环境中指导决策的大脑机制。这些机制是通过跨皮层和皮层下区域的大型神经网络的相互作用来实现的。跟踪单个试验的群体反应动态，并将其与内部认知状态和公开行为联系起来，对于当前决策模型的深入测试至关重要。在这里，我们提出的活动模式的特点大群体的神经元（100+）在前额叶皮质的猕猴从事决策任务。我们将通过开发最简约的概率模型来解释种群中激活的时空模式，该模型考虑了神经元的成对和高阶相互作用。该模型将被用来表征网络的响应流形，并量化其在不同的任务时期的动态。我们的方法是独特的，因为我们建议扩展最大熵框架以直接捕获高维动态，而不是试图使用机器学习工具将种群动态嵌入低维流形中。此外，通过表征细胞之间的功能依赖性，我们可以根据子网络图案和计算来映射大型网络的架构和设计。最后，我们将研究响应的噪声波动或人工操纵网络活动如何影响其动态，并修改或干扰其计算。在我们的建模框架内仔细研究这些实验的结果，有助于解决长期存在的关于决策的问题，包括心理模型的神经基础和初始状态对行为的影响。 我们的工作将在两个领域产生更广泛的影响。首先，我们将绘制的发现功能子网络及其计算的路径将可用于神经科学的各个子领域。我们将大大推进神经科学家可用的数据分析和计算建模工具，因此，将促进未来使用高维神经数据对正常精神功能和精神障碍的研究。第二，表征信息编码和反应动力学在前额叶皮层揭示了决策机制和认知能力的出现在复杂的神经网络。决策缺陷是许多神经和精神疾病的核心，包括精神分裂症、阿尔茨海默氏症和帕金森氏症。针对这些缺陷，已经提出了几种行为和药理学疗法，但我们缺乏对它们在神经系统水平上如何工作的清晰理解。为了开发下一代疗法，我们需要了解认知过程如何在多个功能水平上出现，从单个神经元到大脑区域网络。我们的工作是朝这个方向迈出的一步。它有可能促进我们对精神障碍病理学的理解，并有助于发现更好的治疗方法。

PROJECT SUMMARY The essence of cognition is choice, and to understand choice we need to understand the brain mechanisms that guide decisions in complex settings. These mechanisms are implemented through interactions of large neural networks across cortical and subcortical areas. Tracking population response dynamics on single trials and relating them to internal cognitive states and overt behavior are critical for incisive tests of current models of decision-making. Here we propose to characterize the activity patterns of large populations of neurons (100+) in the prefrontal cortex of macaque monkeys engaged in decision-making tasks. We will explain the spatiotemporal patterns of activation in the population by developing the most parsimonious probabilistic model that takes into account pairwise and higher-order interactions of neurons. The model will be utilized to characterize response manifolds of the network and quantify its dynamics in different task epochs. Our approach is unique because rather than trying to embed the population dynamics in a low dimensional manifold using Machine Learning tools, we propose to extend the Maximum Entropy framework to directly capture the high-dimensional dynamics. Further, by characterizing functional dependencies among cells we can map the architecture and design of large networks in terms of subnetwork motifs and computations. Finally, we will investigate how noisy fluctuations of responses or artificial manipulation of network activity influences its dynamics and modifies or disturbs its computations. Scrutinizing the results of these experiments within our modeling framework makes headway toward addressing long-standing questions about decision-making, including the neural basis of psychological models and effects of initial state on the behavior. Our work will have broader impacts in two domains. First, the path that we will chart for discovering functional subnetworks and their computations will be useable in various subfields of neuroscience. We will significantly advance data analysis and computational modeling tools available to neuroscientists and, therefore, will facilitate future studies of normal mental functions and mental disorders using high-dimensional neural data. Second, characterizing information encoding and response dynamics in the prefrontal cortex sheds light on mechanisms of decision-making and emergence of cognitive abilities in complex neural networks. Deficits of decision-making are at the heart of a number of neurological and psychiatric disorders including schizophrenia, Alzheimer's, and Parkinson's disease. Several behavioral and pharmacological therapies have been proposed for those deficits, but we lack a clear understanding of how they work at the level of neuronal systems. To develop the next generation of therapies, we need to understand how cognitive processes emerge across multiple functional levels, from individual neurons to networks of brain areas. Our work is a step in that direction. It has the potential to advance our understanding of pathology of mental disorders and help with the discovery of better treatments.