Multi-Resolution Partitioning and Kinetic Function Estimation for Dynamic Biochemical Network Model Development

这项跨学科的研究致力于创建生物系统的动态模型，其预测能力超出了当前一代描述性静态模型的能力。一个生物系统，如细胞，可以被概念化为一个复杂的、完整的生化反应网络。代谢反应是一类重要的反应，执行基本的细胞功能，如能量产生、生物合成、有害废物和副产物的消除。细胞代谢预测模型作为基础和应用研究的工具提供了广泛的益处。在公共卫生的背景下，代谢模型可以用来整合关于药物疗效的新的实验室和临床数据，比较健康和患病的组织，并预测正在开发的新药的潜在有害副作用。代谢模型在生物技术中也扮演着重要的角色，因为它们能够设计和优化基因工程微生物细胞，生产工业上有用的大宗和增值化学品(如生物燃料)。本研究重点研究了代谢网络的两种创新建模技术。这两种技术的驱动原理是结构分析和功能分析的集成。第一种技术是多分辨率结构建模，它结合了自上而下的模块化和自下而上的单个反应的功能抽象。第二种技术补偿了数学建模过程中的不完全信息。这项工作研究了在使用噪声数据进行模型校准的同时，对每个主要反应集的反应速率定律函数和相关参数的联合估计。这两种建模技术及其相关算法在从肝细胞(肝细胞)培养收集的实验数据上进行了测试，肝细胞是一个具有代表性的、研究得很好的模型系统，具有生化复杂性，与公共健康相关。通过这项研究获得的建模技术具有一般性，例如，定向和时变相互作用的数学抽象，这不仅适用于代谢网络，也适用于其他类型的重要生化(例如信号、基因调控等)。网络。

This interdisciplinary research investigates creating dynamic models of biological systems with predictive power that is beyond the capabilities of current generation of descriptive, static models. A biological system such as a cell can be conceptualized as a complex integrated network of biochemical reactions. Metabolic reactions are an important class of reactions performing essential cellular functions such as energy generation, biosynthesis, and harmful waste and byproduct elimination. Predictive models of cellular metabolism offer broad benefits as tools for both basic and applied research. In the context of public health, metabolic models can be used to integrate new laboratory and clinical data on drug efficacy, to compare healthy and diseased tissues, and to predict potentially harmful side effects of new drugs under development. Metabolic models also play an essential role in biotechnology as they enable the design and optimization of genetically engineered microbial cells that produce industrially useful bulk and value-added chemicals (e.g. biofuels).This research focuses on two innovative modeling techniques for metabolic networks. The driving principle for both techniques is integration of structural and functional analyses. The first technique is multi-resolution structural modeling, which combines top-down modularization and bottom-up functional abstraction of individual reactions. The second technique compensates for incomplete information during mathematical modeling. This work investigates a co-estimation of reaction rate law functions and relevant parameters for each dominant reaction set while using noisy data for model calibration. The two modeling techniques and their associated algorithms are tested on experimental data collected from cultures of liver cells (hepatocytes), a representative and well-studied model system with biochemical complexity and relevance to public health. The modeling techniques obtained through this study feature generic aspects, e.g. mathematical abstraction of directed and time-varying interactions, which apply not only to metabolic networks but also other types of important biochemical (e.g. signaling, gene regulatory, etc.) networks.