下オリーブ核神経細胞群におけるギャップ結合および抑制性結合のベイズ推定

我们过去研究的主要成果是成功构建了一个复杂的框架，以解决估计下橄榄神经元尖峰序列的间隙连接和抑制性电导的问题（Hoang等人，2015）。传统上，基于多个神经元的峰值时序估计神经元网络模型参数的问题是严重不适定的。因此，逆问题需要一种随机方法，在许多可能的解中找到最可能的解。我们开发了一个贝叶斯估计的理论框架，该框架允许在神经元约束库中参数估计的分段波动，以补偿建模误差。神经元约束避免了我们之前对高度非平稳实验数据的研究（Onizuka et al., 2013）中出现的过度拟合。我们在（Hoang和Tokuda， 2015）中进一步验证了该方法，该方法使用模拟峰值数据作为测试数据。在两个数据集上的结果证实了贝叶斯方法具有克服非平稳动力学问题和解决逆问题的病态性的潜力。因此，我们认为我们提出的贝叶斯框架是评估神经科学中感兴趣的参数的有用工具。

The primary achievement of our past research was a successful construction of a sophisticated framework to resolve the problem of estimating gap-junctional and inhibitory conductance from spike trains of inferior olive neurons (Hoang et al., 2015). Traditionally, the problem to estimate model parameters of a network of neurons based on spike timings of a number of neurons is severely ill-posed. That inverse problem thus needs a stochastic approach which finds most likely solution among many possible ones. We developed a theoretical framework of Bayesian estimation, which allows segmental fluctuations of parameter estimates in the neuronal constraint base, in order to compensate the modeling errors. The neuronal constraint avoids over-fitting as happened in our previous study (Onizuka et al., 2013) for highly non-stationary experimental data. We further validated that method in (Hoang and Tokuda, 2015), which utilized simulated spike data as the test data. The results on the both data sets confirmed that the Bayesian method has the potential to overcome the problem of non-stationary dynamics and resolved the ill-posedness of the inverse problem. We thus argue that our proposed Bayesian framework is a useful tool to evaluate the parameters of interest in neuroscience.