D3SC: EAGER: Data-driven design of molecular models from microscopic dynamics and experimental data

有了这个奖项，化学系的CTMC计划资助莱斯大学的Cecelia Clementi教授开发用于研究各种尺度物质性质的模型。化学科学的一个根本挑战是弥合在原子尺度上研究和操纵物质的能力与在宏观尺度上理解和预测性质的愿望之间的差距。最近，用于实验观测的高通量和高分辨率技术有了巨大的增长。此外，在微观水平上模拟分子系统的高性能技术也有所增加。这些进步导致了大量且不断增加的高维数据。因此，最近对数据分析技术的兴趣激增。特别是从模拟中提取基本特征、集体变量或代表性状态的技术。这些模拟数据必须与实验数据相一致。然而，除了极少数例外，这些调和技术纯粹是描述性的，不允许制定规范宏观行为的一般原则。此外，很难扩大到更大和更复杂的系统。该项目开发了一种新的通用方法来应对这一挑战。这些方法适用于非常不同的化学系统，从细胞中的信号传导、多相催化到聚合物刷的设计。拟议的研究影响了化学，物理和数学的学生和研究人员的大型跨学科社区。该项目负责编制适用于化学系统的计算和数学方法课程。课程开发包括本科和研究生课程。该项目还招募和指导妇女和少数民族本科生和研究生。这项活动是通过与莱斯大学塔皮亚中心合作进行的，该项目正在制定一个总体框架，以获得控制分子系统的有效动力学模型（结构、方程和参数）。关键的假设是，为了能够理解和建模宏观系统，需要使用纯描述性模型来定义来自数据的生成模型。模型是在中尺度从微尺度模拟和多尺度实验数据。该方法从根本上不同于现有的粗粒度技术或模型简化方法。宏观模型的形式以及有效的动力学方程都是从数据中学习的。为了弥合微观和宏观系统之间的差距，生成了可以嵌入高阶模拟的功能构建块。该方法研究了相对一般的组织原则是否以及如何从不同化学体系中的大量原子自由度的相互作用中出现。这种建模方法有可能成为将大量化学数据集成到定量，机械和可理解的模型中的基石。这些模型能够解释不同的分子组分如何组织和相互作用，作为时间和空间的函数，在宏观尺度上执行功能。

With this award the CTMC program in the Division of Chemistry is funding Professor Cecelia Clementi at Rice University to develop models for studying properties of matter at various scales. A fundamental challenge for the chemical sciences is to bridge the gap between the ability to study and manipulate matter at the atomistic scale with the desire to understand and predict properties at a macroscopic scale. Recently, there has been an immense increase in high-throughput and high-resolution technologies for experimental observation. In addition there is an increase in high-performance techniques to simulate molecular systems at a microscopic level. These advances have resulted in a vast and ever-increasing amounts of high-dimensional data. Consequently, there is a recent surge of interest in data analysis techniques. In particular, techniques that extract essential features, collective variables or representative states from simulations. These simulation data have to be reconciles with experimental data. However, with very few exceptions, these reconciling techniques are purely descriptive and do not allow the formulation of general principles regulating the macroscopic behavior. Furthermore it is difficult to scale up towards significantly larger and more complex systems. This project develops a new and general approach to address this challenge. The approaches are applicable to very different chemical systems, ranging from signal transduction in cells, over heterogeneous catalysis to the design of polymer brushes. The proposed research impacts a large interdisciplinary community of students and researchers in Chemistry, Physics and Mathematics. The project undertakes curriculum development in computational and mathematical methods applied to chemical systems. Curriculum development includes undergraduate and graduate courses. The project is also recruiting and mentoring women and minority undergraduate and graduate students. This activity is conducted through a collaboration with the Tapia Center at Rice University.The project is developing a general framework to obtain the effective dynamical models (structure, equations and parameters) governing molecular systems. The key hypothesis is that, in order to be able to understand and model macroscopic systems, there is a need to use purely descriptive models to define generative models from data. Models are developed at the mesoscale from microscale simulations and multiscale experimental data. The approach is fundamentally different from available coarse-graining techniques or model reduction methods. Both the form of the macroscopic model as well as the effective dynamical equations are learned from data. Functional building blocks that can be embedded in higher order simulations are generated in order to bridge the gap between microscopic and macroscopic systems. The method investigates if and how relatively general organizing principles emerge from the interactions of a multitude of atomic degrees of freedoms in different chemical systems. This modeling approach has the potential to serve as a keystone to integrate vast amounts of chemical data into quantitative, mechanistic and comprehensible models. Such models are able to explain how different molecular components organize and interact as a function of time and space in performing functions at the macroscopic scale.