Modeling and Analysis of Neural Codes for Spatial Computation

当我们的大脑在生命早期发育时，几何空间的内在表征开始为建立世界与我们感知的感官世界之间的联系奠定了基础。这包括知道你在哪里，知道如何导航到目标。在神经退行性疾病，如阿尔茨海默氏病（AD），这些设施是一些第一个行为特征恶化。这为使用空间导航作为AD早期诊断中的行为标记打开了大门。空间导航被认为是由认知规划和内在认知地图驱动的。使用虚拟现实（VR），我们可以分离空间导航策略，并开始了解这些功能中的哪一个受影响最大。通过这些VR任务，我们希望训练机器学习算法来学习任务表现和神经退行性缺陷之间的关系。这很可能提供负担得起的，可访问的和无压力的诊断工具。我们建议首先在神经退行性小鼠模型中研究机器学习预测的实用性，然后是各种AD小鼠模型，最后在包括健康，风险和AD患者队列的人类上验证我们的方法。我们还希望开发新的强化学习算法，并借助大脑中空间计算的生物学合理模型。这些模型将成为理解空间计算缺陷的更精细特征和更广泛的神经计算的基础。

As our brains' develop in early life, intrinsic representations of geometric space begin to set the groundwork for establishing connections between the world and the sensory world we perceive. This includes knowing where you are and knowing how to navigate to a target. In neurodegenerative diseases such as Alzheimer's' disease (AD), these facilities are some of the first behavioural traits to deteriorate. This opens the door to using spatial navigation as a behavioural marker in the early diagnosis of AD. Spatial navigation is proposed to be driven by both cognitive planning and intrinsic cognitive maps. Using virtual reality (VR), we can dissociate spatial navigation strategies and begin to understand which of these features is most affected. With these VR task, we hope to train machine learning algorithms to learn the relationships between task performance and neurodegenerative deficits. This may well provide diagnostic tools which are affordable, accessible and stress free.We propose to investigate the utility of machine learning prediction first in neurodegenerative mouse models, followed by various AD mouse models, before finally validating our methods on humans including healthy, at-risk and AD patient cohorts. We further hope to develop new reinforcement learning algorithms aided by biologically plausible models of spatial computation in the brain. These models will form the basis of understanding the finer features of spatial computation deficits and more broadly neural computation in general.