Genomic data assimilation for estimating hybrid functional Petri net from time-course gene expression data.
Genomic data assimilation for estimating hybrid functional Petri net from time-course gene expression data.
复制标题
DOI：
10.11234/gi1990.17.46
复制
发表时间：
2006
期刊：
影响因子：
--
通讯作者：
Masao Nagasaki;R. Yamaguchi;Ryo Yoshida;S. Imoto;Atsushi Doi;Y. Tamada;H. Matsuno;S. Miyano;T. Higuchi
中科院分区：
文献类型：
--
作者：
Masao Nagasaki;R. Yamaguchi;Ryo Yoshida;S. Imoto;Atsushi Doi;Y. Tamada;H. Matsuno;S. Miyano;T. Higuchi
本文提出了一种自动构造混合功能Petri网的方法，并将其作为生物通路的仿真模型。我们考虑的问题是我们如何选择参数的值和我们如何设置网络结构。通常，我们根据经验调整这些未知因素，使模拟结果与生物学知识一致。显然，这种方法在感兴趣网络的大小上有限制。为了扩展模拟模型的能力，我们建议使用最初建立在地球物理模拟科学领域的数据同化方法。我们提供了基因组数据同化框架，通过使用非线性状态空间模型在我们的模拟模型和观察到的数据（如微阵列基因表达数据）之间建立联系。我们的基因组数据同化的一个关键思想是，模拟模型中的未知参数转换为状态空间模型的参数，并获得的估计作为最大后验估计。在参数估计过程中，利用仿真模型在状态空间模型中生成系统模型。这样的配方使我们能够处理的贝叶斯统计推断的框架内的模型构建和参数调整。特别是，贝叶斯方法为我们提供了一种方法来控制过拟合的参数估计，这是必不可少的构建一个可靠的生物途径。我们证明了我们的方法使用合成数据的有效性。因此，使用基因组数据同化的参数估计工作得很好，网络结构被适当地选择。
We propose an automatic construction method of the hybrid functional Petri net as a simulation model of biological pathways. The problems we consider are how we choose the values of parameters and how we set the network structure. Usually, we tune these unknown factors empirically so that the simulation results are consistent with biological knowledge. Obviously, this approach has the limitation in the size of network of interest. To extend the capability of the simulation model, we propose the use of data assimilation approach that was originally established in the field of geophysical simulation science. We provide genomic data assimilation framework that establishes a link between our simulation model and observed data like microarray gene expression data by using a nonlinear state space model. A key idea of our genomic data assimilation is that the unknown parameters in simulation model are converted as the parameter of the state space model and the estimates are obtained as the maximum a posteriori estimators. In the parameter estimation process, the simulation model is used to generate the system model in the state space model. Such a formulation enables us to handle both the model construction and the parameter tuning within a framework of the Bayesian statistical inferences. In particular, the Bayesian approach provides us a way of controlling overfitting during the parameter estimations that is essential for constructing a reliable biological pathway. We demonstrate the effectiveness of our approach using synthetic data. As a result, parameter estimation using genomic data assimilation works very well and the network structure is suitably selected.