Building a Complete, Predictive, Data-Driven Model of Action Selection During Olfactory Navigation

摘要 为了生存，活着的有机体必须收集关于他们环境的信息，并利用这些信息来选择合适的 行为。然而，来自环境的信息往往是嘈杂的、不完整的和模棱两可的。目前， 没有理论或模型能全面解释神经系统是如何解决导航问题的 关于嘈杂的信息。没有这样的理论，我们就不能提高系统的生存能力或自主能力 机器通过处理通常可用的不完美的感觉信息来做出更好的决定 敬他们。 我们建议建立一个完整的数据驱动的模型，说明神经系统如何将嘈杂的感觉信息 进入导航过程中的动作选择。我们以前已经能够通过以下方式破译这一过程的各个方面 研究果蝇幼虫--一种小而透明的生物，特别擅长 尽管只有10,000个神经元，但仍能闻到食物的气味。我的实验室已经开发出方法来 严格量化气味环境；测量神经元如何代表这些气味；自动跟踪幼虫 运动；为幼虫创造虚拟感官现实；以及改变幼虫在- 对光遗传学的需求。我们最近还绘制了幼虫神经内的一条完整路径 系统。在这里，我们将确定嘈杂的感官信息如何以及何时导致幼虫重新定向(停止 和转弯)，因为它是导航到一个有吸引力的气味来源(趋化性)。我们的目标是发现 积累、过滤和处理嘈杂的感觉证据并使用模棱两可的神经机制 做出连贯的感性决策的信息(行动选择)。通过将理论、实验、 和建模，我们将迭代地构建一个预测细胞和电路级别的量化模型 将感觉(嗅觉)信号转换为导航决策(趋化性)的计算，即 对环境干扰(噪声)有很强的抵抗力。

Abstract To survive, living organisms must collect information about their environment and use it to select appropriate behaviors. However, information from the environment is often noisy, incomplete and ambiguous. Currently, no theory or model comprehensively explains how nervous systems solve the problem of navigation based on noisy information. Without such a theory, we cannot improve the ability of living systems or autonomous machines to make better decisions by processing the imperfect sensory information that is typically available to them. We propose to build a complete data-driven model of how nervous systems turn noisy sensory information into action selection during navigation. We have previously been able to decipher aspects of this process by studying the Drosophila melanogaster larva — a small, transparent organism that is exceptionally good at navigating towards food odors despite having only 10,000 neurons. My lab has developed methods to rigorously quantify odor landscapes; measure how neurons represent these odors; automatically track larval movement; create virtual sensory realities for the larva; and change the real-time behavior of the larva on- demand with optogenetics. We have also recently mapped an entire pathway within the larval nervous system. Here, we will determine how and when noisy sensory information causes the larva to reorient (stop and turn) as it is navigating towards an attractive odor source (chemotaxis). Our objective is to uncover the neural mechanisms that accumulate, filter, and process noisy sensory evidence and use ambiguous information to make coherent perceptual decisions (action selection). By combining theory, experiments, and modeling, we will iteratively build a quantitative model which predicts the cellular and circuit-level computations transforming sensory (olfactory) signals into navigational decision-making (chemotaxis) that is robust to environmental disturbances (noise).