DDDAS-SMRP: Development of a closed-loop identification machine for bionetworks (CLIMB) and its application to nucleotide metabolism

细胞代谢的化学反应网络是生物体最保守的特征之一。它也是疾病治疗的关键靶点，有几种重要的药物可以抑制DNA和RNA前体的生物合成(核苷酸代谢)。系统生物学、分析化学和计算机科学的最新进展为开始构建生物网络集成行为的定量、动态模型提供了机会，如核苷酸代谢。然而，在给定非线性网络行为和有限的实验室测量数量的情况下，获得准确的动态模型的主要障碍在于从噪声实验数据中提取定量模型的难度。这项拟议的研究旨在开发一种用于分子生物网络的闭环识别机(SCRIBE)，通过反复几轮计算和计算指导的实验，实现可靠和经济高效的模型识别。GRAPH包括两个核心计算组件，(1)控制模块，其设计用于应用于目标网络的优化的化学通量以用于参数识别；(2)逆模块，其基于实验结果计算可行的参数值。与传统自适应辨识技术不同的特点包括利用非线性全局逆算法提取与实验测量一致的模型参数分布，并在非线性闭环系统学习控制算法中利用该参数分布来指导进一步的实验。该技术将被充分开发和应用，以实现对大肠杆菌和酿酒酵母核苷酸代谢的定量了解。攀登设计的化学物质将被引入到营养有限的、稳定的活细胞培养中。代谢网络的动态响应将通过分析细胞内核苷酸含量在应用攀登设计的化学通量期间和之后进行监测。拟议研究的智力价值在于应用迭代的计算和实验来处理对传统方法来说太难的网络识别挑战。它还在于设计适合于高度非线性网络和有限、噪声测量的现实世界挑战的计算模块。最后，这一努力的最终结果将是计算与生长细胞的培养、收集和代谢组测量的简化集成。

The chemical reaction network of cellular metabolism is one of the most highly conserved features of organisms. It is also a key target of disease treatment, with several important drugs inhibiting the biosynthesis of DNS and RNA precursors (nucleotide metabolism). Recent advances in systems biology, analytical chemistry, and computer science provide the opportunity to begin construction quantitative, dynamic models of the integrated behavior of bionetworks like nucleotide metabolism. A major barrier to obtaining accurate dynamic models, however, lies in the difficulty of extracting the quantitative models from noisy experimental data, given nonlinear network behavior and the limited number of laboratory measurements. The proposed research aims to develop a closed-loop identification machine for molecular bionetworks (CLIMB) to enable reliable and cost-effective model identification through iterative rounds of computation and computation-guided experimentation. CLIMB involves two core computational components, (1) a control module which designs optimized chemical fluxes for application to the target network for the purpose of parameter identification, and (2) an inversion module which calculates feasible parameter values based on the experimental results. The features that distinguish CLIMB from traditional adaptive identification techniques include the utilization of a non linear, global inversion algorithm to extract the distribution of model parameters consistent with the experimental measurements, and the utilization of this parameter distribution in a nonlinear, closed-loop learning control algorithm to guide further experiments. The CLIMB technique will be fully developed and applied to achieve a quantitative understanding of the nucleotide metabolism of E. coli and S. cerevisiae. CLIMB-designed chemical influxes will be introduced to the nutrient-limited, steady-state cultures of living cells. The dynamic response of the metabolic network will be monitored by analyzing intracellular nucleotide content during and after application of the CLIMB-designed chemical flux. The intellectual merit of the proposed research lies in applying iterative rounds of computation and experimentation to handle network identification challenges too difficult for classical approaches. It further lies in designing computational modules suitable for the real-world challenges of highly nonlinear networks and limited, noisy measurements. Finally, the effort will culminate in streamlined integration of computation with culture, collection, and metabolomic measurement of growing cells.