从脉冲序列推断单突触的因果推断方法
A causal inference approach of monosynapses from spike trains
AI总结：
本文提出一种从脉冲序列推断成对单突触效应的严格因果推断框架，通过时间尺度分离识别因果交互并开发快速算法，同时结合动力系统模拟和光遗传学实验探讨生物物理参数与统计因果定义之间的联系及实验假设检验。
摘要：
自20世纪60年代以来，神经科学家一直致力于从同步记录的脉冲序列中估计突触特性（如连接性和强度）的问题。近年来，随着高密度神经记录和光遗传学技术的快速发展，这一问题重新引起了人们的兴趣，这些技术可用于校准关于功能连接的因果假设。本文建立了一个严格的因果推断框架，用于分析脉冲序列之间成对的兴奋性和抑制性单突触效应。通过按时间尺度分离成对脉冲序列中的脉冲相互作用来识别因果交互。同时，本文还开发了用于计算相关量准确估计值的快速算法。通过该框架的视角，本文在一系列动力系统模拟中考察了生物物理参数与脉冲序列之间因果关系的统计定义之间的联系。在理想化设置下，我们证明了本文建立的突触因果度量与Tian和Pearl提出的因果概率之间存在对应关系。由于因果概率是在不同的假设下推导出来的，并且包含来自实验随机化的数据，这为利用近细胞或光遗传学刺激检验突触推断框架的假设开辟了可能性。我们使用生物物理上详细的通道视紫红质模型模拟了这样一个实验，并表明随机化并未实现；即使在强刺激下，强混淆仍然存在。一个主要目标是探究如何通过精心表述的因果假设更好地指导神经刺激实验的设计，并反过来支持对这些假设的实验检验。
英文摘要：
Neuroscientists have worked on the problem of estimating synaptic properties, such as connectivity and strength, from simultaneously recorded spike trains since the 1960s. Recent years have seen renewed interest in the problem, coinciding with rapid advances in the technology of high-density neural recordings and optogenetics, which can be used to calibrate causal hypotheses about functional connectivity. Here, a rigorous causal inference framework for pairwise excitatory and inhibitory monosynaptic effects between spike trains is developed. Causal interactions are identified by separating spike interactions in pairwise spike trains by their timescales. Fast algorithms for computing accurate estimates of associated quantities are also developed. Through the lens of this framework, the link between biophysical parameters and statistical definitions of causality between spike trains is examined across a spectrum of dynamical systems simulations. In an idealized setting, we demonstrate a correspondence between the synaptic causal metric developed here and the probabilities of causation developed by Tian and Pearl. Since the probabilities of causation are derived under distinct assumptions and include data from experimental randomization, this opens up the possibility of testing the synaptic inference framework's assumptions with juxtacellular or optogenetic stimulation. We simulate such an experiment with a biophysically detailed channelrhodopsin model and show that randomization is not achieved; strong confounding persists even with strong stimulations. A principal goal is to ask how carefully articulated causal assumptions might better inform the design of neural stimulation experiments and, in turn, support experimental tests of those assumptions.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
