体外培养大鼠皮质神经网络中功能性连接结构辨识研究

探明网络中动态信息流的传导路径和运输方向，从而揭示大脑的生理和病理机制是神经科学研究的重要目标之一。本课题针对生物神经元网络中功能性连接结构辨识问题，发展了一种基于随机重启爬山的动态贝叶斯网络优化算法，该方法能够仅利用多电极阵列获得的多通道电脉冲序列数据，辨识出生物神经元网络中功能性连接。其基本思想为：计算每一候选网络结构的贝叶斯信息准则评分。该评分兼顾网络结构似然度函数和网络复杂度两种因素的影响，具有最小评分的网络结构即被选定为最优结构。动态贝叶斯网络方法本质上为一种概率方法，能够克服以往生物网络连接结构辨识法大都依赖于线性分析的局限性，不再受具体支路连接形式的限制，特别适用于大规模生物网络分析。最后，将所提出的方法应用于体外培养的大鼠脑皮质神经元网络产生的脉冲序列数据，辨识其中功能性连接的存在情况、性质与分布特点，从系统水平阐明生物神经网络中支配其特定网络行为的功能性连接结构特点。

One of major targets of neural science researches is to reveal biophysical and pathogenic mechanism of brains through investigating the transduction pathway and transportation direction of dynamical information flows in BNNs (Biological Neuronal Network). We shall propose a novel method, called DBNOA (Dynamic Bayesian Network Optimization Algorithm), based on a RRHC (Random Restart Hill Climbing), by which the functional connections embedded in BNNs can be identified using the spike train data measured by MEA (Multi-electrode Array) devices. The basic idea of DBNOA is to calculate a BIC (Bayesian Information Criterion) score for each candidate network structure. Then, a network structure with the minimum loss score, which compromises network structure likelihood fitting network dynamical behaviors and the network complexity, is elected as the optimal network connective structure using RRHC. Because of essentially being a probability method, DBNOA avoids the limitations of traditional linear analysis methods and is free of specific network connective forms. Thus it is especially suitable for large-scale network analysis. Finally, the proposed identification method will be applied to in vitro cultured rat cortical neural networks for elucidating relationship between their specific network behaviors and the existence, the property, and the distribution of underlying functional connective structures.