An Efficient Data Assimilation Schema for Restoration and Extension of Gene Regulatory Networks Using Time-Course Observation Data
An Efficient Data Assimilation Schema for Restoration and Extension of Gene Regulatory Networks Using Time-Course Observation Data
复制标题
DOI：
10.1089/cmb.2014.0171
复制
发表时间：
2014-11-01
影响因子：
1.7
通讯作者：
Akutsu, Tatsuya
中科院分区：
文献类型：
--
作者：
Hasegawa, Takanori;Mori, Tomoya;Akutsu, Tatsuya
基因调控网络(GRN)在维持细胞内复杂的生物系统中起着核心作用。虽然我们可以通过整合文献中记录的生物相互作用来构建GRN，但它们可能包括可疑数据和缺乏信息。因此，迫切需要一种方法来评估所构建的网络的有效性；基于模拟的方法已经被应用，其中生物观测数据被同化。然而，这些方法应用的非线性模型需要很高的计算能力来评估一个只由几个基因组成的网络。因此，为了通过修改和扩展基于文献的GRN来寻找其仿真模型能够更好地预测数据的候选网络，迫切需要一种高效且通用的方法。我们应用一个能代表基因组合调控效应的组合转录模型作为生物模拟模型，在状态空间模型中再现基因表达的动态行为。在该模型下，应用无迹卡尔曼滤波得到隐含状态的近似后验概率分布，通过EM算法对参数值进行有效估计，最大化对观测数据的预测能力。利用这种方法，我们提出了一种新的算法来修正文献中报道的GRN，使其仿真模型与观测数据相一致。通过与已有的合成网络方法的对比分析，验证了该方法的有效性。最后，应用该方法对基于京都基因与基因组百科全书(KEGG)的酵母细胞周期网络进行了扩展，增加了候选基因，以更好地预测真实的mRNA表达数据。
Gene regulatory networks (GRNs) play a central role in sustaining complex biological systems in cells. Although we can construct GRNs by integrating biological interactions that have been recorded in literature, they can include suspicious data and a lack of information. Therefore, there has been an urgent need for an approach by which the validity of constructed networks can be evaluated; simulation-based methods have been applied in which biological observational data are assimilated. However, these methods apply nonlinear models that require high computational power to evaluate even one network consisting of only several genes. Therefore, to explore candidate networks whose simulation models can better predict the data by modifying and extending literature-based GRNs, an efficient and versatile method is urgently required. We applied a combinatorial transcription model, which can represent combinatorial regulatory effects of genes, as a biological simulation model, to reproduce the dynamic behavior of gene expressions within a state space model. Under the model, we applied the unscented Kalman filter to obtain the approximate posterior probability distribution of the hidden state to efficiently estimate parameter values maximizing prediction ability for observational data by the EM-algorithm. Utilizing the method, we propose a novel algorithm to modify GRNs reported in the literature so that their simulation models become consistent with observed data. The effectiveness of our approach was validated through comparison analysis to the previous methods using synthetic networks. Finally, as an application example, a Kyoto Encyclopedia of Genes and Genomes (KEGG)-based yeast cell cycle network was extended with additional candidate genes to better predict the real mRNA expressions data using the proposed method.