CAREER: Data-Driven Systematic Hierarchical Modeling

佛蒙特州立大学农学院的Jianing Li由化学系的化学理论，模型和计算方法计划以及刺激竞争研究的既定计划（EPSCoR）共同支持，使用计算机模拟模型预测材料的特性。如何将分子结构与材料性质联系起来的问题一直是化学科学的中心问题。计算机建模在帮助回答这个基本问题方面具有非常宝贵的价值，但预测更大结构的性质仍然非常困难。目前的仿真方法往往达到其局限性，因为它们不能仿真足够长的时间的大型系统。李博士现在正在利用以前完成的分子模拟的大量数据。她正在发明一种有效的方法，可以从现有数据中自动学习，以建立新的分子模型。这些模型能够降低模拟大量潜在分子组分的难度。通过将这些模型与材料特性（如稳定性、形状和尺寸）的预测相联系，李博士正在使用模拟方法来筛选天然和人造聚合物的理想特性，并加速新材料的发现。在这个项目中设计的材料将是生物相容性，生物活性纳米材料在传感，药物输送，组织工程等众多应用该项目还提供了教育机会，为学生在他们的职业发展的多个阶段，通过培养研究生与跨学科的重点，以及鼓励本科生早期独立实验与分子建模。 教育活动将通过为本科生提供新的学习和研究机会来扩大STEM的参与。 将设立旅行奖励，以鼓励佛蒙特州代表性不足的学生与他们的导师和同龄人一起参加绿色山冬令营。对于未来软材料的技术进步（例如，创建由肽，DNA和有机聚合物形成的新的可编程，生物相容性纳米结构），了解复杂的自组装过程并能够准确预测所得结构至关重要。分层模型是理解这些过程的宝贵工具，因为它可以检查和预测（通常比实验更详细）自组装如何在相关的原子，纳米和介观尺度上发生。然而，发明更强大的分层计算方法的未来，通用的高度粗粒度（HCG）的力场共轭与有效的映射方法是必要的。为了应对这些挑战，Li博士正在创建一种用于多尺度建模的系统化数据驱动层次（STAIR）方法。STAIR旨在克服目前可用的系统粗粒化方法的主要缺点，这些方法仍然需要大量的人力专业知识和劳动力来开发力场。具体而言，STAIR通过创新的神经网络算法取代了昂贵的拟合过程，并减少了颗粒类型确定，特别校正等任务中的人工努力。长期目标是从相对简单的构建模块指导复杂纳米材料的合理设计，总体目标是发明分层，自适应建模方法，以指导肽和DNA为基础的自我，该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识产权进行评估来支持。优点和更广泛的影响审查标准。

Jianing Li of the University of Vermont & State Agricultural College is jointly supported by the Chemical Theory, Models and Computational Methods program in the Division of Chemistry and the Established Program to Stimulate Competitive Research (EPSCoR) to predict the properties of materials using computer simulation models. The question of how to correlate molecular structures to material properties has been central to the chemical sciences. Computer modeling has been invaluable to help answer this fundamental question, but it is still very difficult to predict the properties of larger structures. Current simulation methods often reach their limitations, since they are not able to emulate large systems for long enough times. Dr. Li is now taking advantage of the immense data from molecular simulations previously completed. She is inventing an efficient approach to automatically learn from existing data to build new molecular models. These models are able to decrease the difficulty of simulating the large amount of underlying molecular components. By connecting these models to predictions of material properties (like stability, shape, and size), Dr. Li is using simulation methods to screen natural and man-made polymers for desirable properties and to accelerate the discovery of new materials. The materials designed in this project will be biocompatible, bioactive nanomaterials for numerous applications in sensing, drug delivery, tissue engineering, etc. The project also provides educational opportunities for students at multiple stages of their career development, by training graduate students with an interdisciplinary focus, as well as by encouraging undergraduate students early on to experiment independently with molecular modeling. The educational activities will broaden STEM participation by providing new learning and research opportunities to undergraduate students. Travel awards will be established to encourage underrepresented students in Vermont to attend the Green Mountain Winter Camp alongside their mentors and peers. For future technological advances in soft materials (e.g. to create new programmable, biocompatible nanostructures formed by peptides, DNAs, and organic polymers) it is critical to understand complex self-assembly processes and to be able to accurately predict the resulting structures. Hierarchical modeling represents an invaluable tool to understand such processes, since it can examine and predict (often in greater detail than experiments) how self-assembly occurs at the relevant atomic, nanoscopic, and mesoscopic scales. However, to invent more powerful hierarchical computational methods for the future, universal highly coarse-grained (HCG) force fields in conjugation with effective backmapping methods are needed. To target these challenges, Dr. Li is creating a systematic, data-driven hierarchical (STAIR) methodology for multiscale modeling. STAIR is designed to overcome major drawbacks of currently available methods for systematic coarse graining, which still require substantial human expertise and labor for force field development. Specifically, STAIR replaces expensive fitting processes by innovative neural network algorithms and reduce human efforts in tasks like particle type determination, ad hoc corrections, etc. With the long-term goal to guide the rational design of complex nanomaterials from relatively simple building blocks, the overall objective is to invent hierarchical, adaptive modeling methods to guide the development of peptide- and DNA-based self-assembled nanomaterials.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.