【作者】 张律文；

【导师】 张武；

【作者基本信息】 上海大学 ， 计算机应用技术， 2010， 博士

【摘要】 随着计算机科学和生物实验技术的发展,通过研究和应用高效算法、结合高性能计算技术,对海量生物学数据进行整合和分析的计算系统生物学正成为21世纪自然科学的核心领域之一。如何分析、阐明和理解基因表达数据从而构建基因调控网络是目前计算系统生物学研究中面临的重大挑战。本论文将计算机科学应用于基因组学,从计算系统生物学的角度研究基因调控网络。论文围绕基因调控网络的构建问题,从数值研究的角度出发,一方面基于基因表达的实验数据,定性地分析基因调控关系;另一方面通过网络的拓扑特性,分析网络的动力学稳定性,以计算模拟的方法生成人工基因调控网络。论文还进一步应用以上算法及其并行思想,构建小鼠脑部神经干细胞(NSC)的基因调控网络,预测重要的调控信号通路,并获取关键基因的功能信息。本论文的创新性工作主要有四个方面:(一)以单基因扰动的稳态数据为分析对象,本文提出了基于线性微分方程模型的逐步回归网络推断算法(Stepwise Network Inference Method,SWNI)。针对基因调控网络较稀疏的特点,根据严格的选元准则,为目标基因逐步选择最具显著影响的调控子子集,克服了高维小样本的实验数据问题,避免了以往算法中对目标基因强制设定最大连接数的不合理性。实验表明,随着网络规模的增大和网络稀疏度的增加,SWNI算法在计算精度、计算效率上均优于同类算法。(二)基于复杂网络理论和真实基因调控网络的拓扑特性,充分考虑网络的生物学鲁棒性和动力学稳定性,本文提出了人工基因网络的生成过程和模拟方法,并构建基因调控网络模拟器(Gene Network Simulator,GN-Simulator)。实验结果表明,GN-Simulator能模拟在拓扑结构上和动力学性质上与真实基因调控网络高度相似的大规模人工网络,并产生用于无偏验证的多样化人工基因表达数据集,为基因调控网络的构建算法提供高效、合理的计算实验平台。(三)论文从基因组学的角度研究基因dcf1在小鼠神经干细胞分化中的作用机制,应用SWNI算法,将数值实验与生物实验相结合,将公共数据库中的基因芯片数据与实验室数据相整合,构建关于dcf1的小鼠神经干细胞基因调控网络。论文预测了pou6f1显著影响脑部神经干细胞分化的功能,该功能与对dcf1的部分研究结果相符;预测了pou6f1对dcf1的显著调控作用,预测结果与已发表的文献中对含POU结构域的蛋白的研究相吻合;分析了dcf1在预测网络中所处的位置,提示dcf1可能处于其调控通路的下游,为进一步探索dcf1的功能奠定了很好的基础。(四)论文将并行计算思想应用于大规模的基因调控网络构建和模拟问题,提高了网络构建的效率,初步满足了对大规模基因表达数据进行分析和处理的需要;通过并行化策略,论文实现了对小鼠神经干细胞高通量基因芯片数据的整合分析,极大提高了处理计算系统生物学中实际问题的能力。更多还原

【Abstract】 Computational systems biology mainly deals with huge quantities of biological data by means of integrating and performing analysis with efficient algorithms and supercomputers. Along with the development of computer science and experimental technique of biology, computational systems biology has become one of the kernel scientific research areas in the 21st century. Comprehensive gene expression analysis and gene regulatory network (GNR) construction are addressed as grand challenges of computational systems biology nowadays. This dissertation contributes an application of computing science to genomics research with the aim to study GRN from the perspective of computational systems biology.Numerical methods which focus on GRN construction are proposed. On one hand, qualitative analysis on gene regulations from expression data is performed. On the other hand, simulation method of generating artificial GRNs is developed by considering specific network topology and dynamic stability. These methods are applied in GRN construction of mouse neural stem cell to predict important regulatory signal pathways and functions of key genes.The innovative results in the four aspects are described below:1. Stepwise Network Inference Method (SWNI) based on linear differential model is proposed to reconstruct sparse GRN from steady state data response to single gene perturbation. A regression subset-selection strategy is adopted to choose significant regulators for a given gene. Then it solves the small size problem for high-dimensional data in order to remove unreasonable limitation of maximum number of regulators by strict selection rules. The numerical experiments indcate that the SWNI is efficient, and outperforms the other methods with the increase in both network size and sparsity.2. Based on complex network theory, gene regulatory network simulation method (GN-Simulator) is developed. Study on the topology of real gene networks, the simulation method for generating artificial gene network is proposed according to the robust biological mechanism and dynamic stability. Numerical experiments demonstrate that large-scale artificial gene networks can be simulated with similar dynamics as real ones, and various synthetic gene expression data can also be generated by the GN-Simulator to provide efficient and reasonable estimation platform for algorithms performance assessment.3. The regulatory mechanism of mouse neural stem cell differentiation by dcf1 is explored on the genome-level. Based on the SWNI, numerical method combined with biological experiment is designed to construct GRN of mouse neural stem cell from public data integrated with laboratory data. Computing-based predictions that pou6f1 might significantly affect the differentiation of neural stem cells in mouse brain and play an important role in regulation of dcf1, are consistent with some findings of dcf1 and research of the POU-domain in the literature. Moreover, the predicted network lays a solid foundation for further exploration of dcf1 function by indicating that dcf1 locates in the downstream of the signaling pathway.4. Parallel computing is considered to construct and simulate the large-scale GRNs, thereby increase efficiency in network inference and meet the computing requirements for large-scale gene expression management preliminarily. With the parallel strategy, synthetical analysis is performed successfully on high through-put gene chip data of mouse neural stem cell and similar issues in computational systems biology will be solved effectively.更多还原