ITR: Statistical Mechanics of Sloppy Models: From Signal Transduction in the Cell Cycle to Forest Modeling and the Nitrogen Cycle

这笔赠款是对提交给信息技术研究(ITR)倡议的一项小提案的回应。PI将利用信息技术和无序系统统计力学的尖端方法来系统地探索大型、复杂的生物模型。这些模型是草率的，关键的挑战是从中提取可靠和可证伪的预测。PI将不会将统计力学应用于物理系统，而是应用于该系统的动力学模型中的参数：因此，这种元建模技术从与当前可用数据一致的整个模型集合中提取预测。元模型将应用于四个问题。其中三个是在细胞信号转导领域研究的系统：PC12细胞中的ERK系统；生长因子受体运输和CDC42；以及结肠癌细胞系Caco-2中的细胞周期。最终的应用是生态学：森林生态系统中的养分循环。PI通过利用信息技术革命带来的计算能力，绕过了数理统计中的传统分析方法。采样方法直接、灵活、易于实现；该方法适当地对多个局部最优状态进行平均，并适当地处理建模过程中的非线性。PI将使用从材料模拟中提取的许多技术想法。通过使用软模对称破缺场来惩罚计算机时间，以及通过使用为加速玻璃材料的模拟而开发的算法，提高了数值效率。当比较不同的胞元类型时，将使用模型的弱耦合复制品来避免由于软模而引起的问题：当比较同一系统的不同理论时，将研究参数的重整化。这些方法在当前的大型计算机上可以高效地并行运行；集合生成可以在网络上启用网格，并在网络上非局部化。元建模的概念将建模过程抽象为统计力学问题，在复杂、无序的系统中将乏味的、不适定的搜索转化为令人兴奋的新问题。从应用的角度来看，来自材料物理界的见解形成了强大的工具，在以前只能进行定性探索的情况下，允许真正的预测能力。从物理学的角度来看，参数空间中软模的流行提供了新的见解和新的方法。从信息技术的角度来看，现在可以对一大类对科学和社会具有重要意义的新问题进行密集的计算分析。正在研究的三个信号转导网络对癌症药物治疗的发展特别感兴趣，并且是对生物信息学和蛋白质组学革命在未来十年将提出的各种挑战的深入研究的预告。对生态元模型的扩展将导致在许多其他复杂系统中的应用，在这些系统中，不完整或初步的数据仍然需要严格的分析。%这项拨款是对提交给信息技术研究(ITR)倡议的一项小提案的回应。PI将利用信息技术和无序系统统计力学的尖端方法来系统地探索大型、复杂的生物模型。这些模型是草率的，关键的挑战是从中提取可靠和可证伪的预测。PI将不会将统计力学应用于物理系统，而是应用于该系统的动力学模型中的参数：因此，这种元建模技术从与当前可用数据一致的整个模型集合中提取预测。元模型将应用于四个问题。其中三个是在细胞信号转导领域研究的系统：PC12细胞中的ERK系统；生长因子受体运输和CDC42；以及结肠癌细胞系Caco-2中的细胞周期。最终的应用是生态学：森林生态系统中的养分循环。

This grant is made in response to a small proposal submitted to the Information Technology Research (ITR) Initiative. The PI will leverage information technology and sophisticated methods from the statistical mechanics of disordered systems to systematically explore large, complex biological models. These models are sloppy, and the key challenge is to extract reliable and falsifiable predictions from them. The PI will apply statistical mechanics not to the physical system, but to the parameters in the dynamical models of the system: this meta-modeling technique thus draws predictions from the entire ensemble of models that are consistent with the currently available data. Meta-modeling will be applied to four problems. Three of these are systems studied in the field of cellular signal transduction: the Erk system in PC12 cells; growth factor receptor trafficking and Cdc42; and the cell cycle in the colon cell line Caco-2. The final application is in ecology: the nutrient cycles in forested ecosystems. The PI bypasses traditional analytical methods in mathematical statistics, by leveraging the computational power made possible by the information technology revolution. The sampling methods are direct, flexible, and easily implemented; the methods properly average over multiple locally optimal states and properly cope with nonlinearity in the modeling process. The PI will use many technical ideas drawn from materials simulations. Numerical efficiency is enhanced through the use of a soft-mode symmetry-breaking field to penalize computer time, and through the use of algorithms developed to accelerate simulations of glassy materials. Weakly-coupled replicas of the model will be used to avoid problems due to soft modes when comparing different cell types: renormalization of parameters will be studied when comparing different theories of the same system. These methods are efficiently run in parallel on current large-scale computers; ensembel genration could be Grid enabled, delocalized over the Web.The meta-modeling concept of abstracting the modeling process into a statistical mechanics problem transforms a tedious, ill-posed search into an exciting new problem in complex, disordered systems. From an applications view, insights drawn from the materials physics community form powerful tools, allowing real predictive power where only qualitative exploration was previously possible. From a physics view, the prevalence of soft modes in parameter space provides new insights and provokes new methods. From an information technology view, intensive computational analysis can now be brought to bear on a large new class of problems of importance to science and society.The three signal transduction networks being studied are of particular interest to the development of drug therapies in cancer, and are a well-studied preview of the kinds of challenges the bioinformatics and proteomics revolution will present in the coming decade. The extension to ecological meta-modeling will lead to applications in many other complex systems where incomplete or preliminary data nonetheless need rigorous analysis.%%% This grant is made in response to a small proposal submitted to the Information Technology Research (ITR) Initiative. The PI will leverage information technology and sophisticated methods from the statistical mechanics of disordered systems to systematically explore large, complex biological models. These models are sloppy, and the key challenge is to extract reliable and falsifiable predictions from them. The PI will apply statistical mechanics not to the physical system, but to the parameters in the dynamical models of the system: this meta-modeling technique thus draws predictions from the entire ensemble of models that are consistent with the currently available data. Meta-modeling will be applied to four problems. Three of these are systems studied in the field of cellular signal transduction: the Erk system in PC12 cells; growth factor receptor trafficking and Cdc42; and the cell cycle in the colon cell line Caco-2. The final application is in ecology: the nutrient cycles in forested ecosystems.***