CAREER: Building an Advanced Cyberinfrastructure for the Data-Driven Design of Chemical Systems and the Exploration of Chemical Space

化学和材料方面的创新是经济发展、繁荣和生活水平提高的关键驱动力。它还为塑造我们社会的能源、环境可持续性和资源等紧迫问题提供了解决方案。这项研究计划旨在通过改变创造潜在创新的过程来提高化学界应对这些挑战的能力。这项研究推动了从试错式搜索转向理性设计的转变。这些研究通过将机器学习等工具引入化学领域，将传统化学研究与现代数据科学相结合。该项目通过建立一个网络基础设施，使数据驱动的研究成为化学界可行和广泛获得的命题，从而使这一新兴领域得以实现和推进，从而成为化学企业不可分割的一部分。这项研究中开发的工具和方法为大规模探索化学空间和更好地理解决定复杂化学系统行为的隐藏机制提供了手段。这些见解可能会加速、简化并最终改变化学开发过程。该项目还解决了使教育适应这一新研究格局的随之而来的需求，以便为下一代科学家和工程师提供充分的装备，为未来的尖端研发培养一支称职和熟练的劳动力队伍，并确保美国学生在国际就业市场上的竞争力。通过促进少数群体参与这一前景光明的领域，它有助于在我们的社会中持续推动平等机会。因此，这个项目促进了科学的进步，促进了NSF的使命所规定的繁荣和福利。尽管人们越来越多地同意数据驱动的发现和合理设计的价值，但这种方法仍然远未成为化学界日常研究的支柱。这项工作解决了三个关键障碍：(I)由于缺乏可用和可获得的工具，数据驱动的研究超出了大多数化学家的范围和范围；(Ii)关于如何使数据科学为化学研究服务的许多基本和实际问题仍未解决；(Iii)数据科学不是化学家正规培训的一部分，因此社区中的许多人缺乏必要的经验和专业知识来利用它。这项研究的中心是创建一个开放的通用软件生态系统，该生态系统将计算机建模、虚拟高通量筛选和大数据分析(即，使用机器学习、信息学和数据库技术对结果数据集进行验证、挖掘和建模)融合到一个集成的研究基础设施中。一个关键的考虑因素是使这个生态系统尽可能全面、健壮和用户友好，这样感兴趣的研究人员就可以随时使用它，而不需要广泛的专业知识。它还作为底层方法、算法和协议创新的开发平台和试验台，也就是说，它允许社区系统和有效地评估不同技术的实用性和性能，包括作为该项目的一部分引入的新技术。正在开发一种元机器学习方法，以建立为网络基础设施提供附加值的准则和最佳做法。这项工作是由具体的分子设计问题推动的，这些问题证明了整体方法的有效性。由数据驱动的研究的质量新颖性及其固有的跨学科所引起的教育挑战，通过利用计算和数据使能的科学与工程的新的研究生课程进行交叉课程和课程开发，创建交互式教材，以及技能建设黑客松倡议来解决。该奖项是与化学、化学理论、模型和计算方法计划部门联合颁发的。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Innovation in chemistry and materials is a key driver of economic development, prosperity, and a rising standard of living. It also offers solutions to pressing problems on energy, environmental sustainability, and resources that shape our society. This research program is designed to boost the chemistry community's capacity to address these challenges by transforming the process that creates underlying innovation. The research promotes a shift away from trial-and-error searches and towards rational design. These combine traditional chemical research with modern data science by introducing tools such as machine learning into the chemical context. This project enables and advances this emerging field by building a cyberinfrastructure that makes data-driven research a viable and widely accessible proposition for the chemistry community, and thereby an integral part of the chemical enterprise. Tools and methods developed in this research provide the means for the large-scale exploration of chemical space and for a better understanding of the hidden mechanisms that determine the behavior of complex chemical systems. These insights can potentially accelerate, streamline, and ultimately transform the chemical development process. The project also tackles the concomitant need to adapt education to this new research landscape in order to adequately equip the next generation of scientists and engineers, to build a competent and skilled workforce for the cutting-edge R&D of the future, and to ensure the competitiveness of US students in the international job market. By promoting minority participation in this promising field, it contributes to a sustained push towards equal opportunity in our society. This project thus promotes the progress of science and advances prosperity and welfare as stated by NSF's mission. While there is growing agreement on the value of data-driven discovery and rational design, this approach is still far from being a mainstay of everyday research in the chemistry community. This work addresses three key obstacles: (i) data-driven research is beyond the scope and reach of most chemists due to a lack of available and accessible tools, (ii) many fundamental and practical questions on how to make data science work for chemical research remain unresolved, and (iii) data science is not part of the formal training of chemists, and much of the community thus lacks the necessary experience and expertise to utilize it. This research centers around the creation of an open, general-purpose software ecosystem that fuses in silico modeling, virtual high-throughput screening, and big data analytics (i.e., the use of machine learning, informatics, and database technology for the validation, mining, and modeling of resulting data sets) into an integrated research infrastructure. A key consideration is to make this ecosystem as comprehensive, robust, and user-friendly as possible, so that it can readily be employed by interested researchers without the need for extensive expert knowledge. It also serves as a development platform and testbed for innovation in the underlying methods, algorithms, and protocols, i.e., it allows the community to systematically and efficiently evaluate the utility and performance of different techniques, including new ones that are being introduced as part of this project. A meta machine learning approach is being developed to establish guidelines and best practices that provide added value to the cyberinfrastructure. The work is driven by concrete molecular design problems, which serve to demonstrate the efficacy of the overall approach. The educational challenges that arise from the qualitative novelty of data-driven research and its inherent interdisciplinarity are addressesed by leveraging a new graduate program in Computational and Data-Enabled Science and Engineering for cross-cutting course and curricular developments, the creation of interactive teaching materials, and a skill-building hackathon initiative. This award is jointly made with the Division of Chemistry's, Chemical Theory, Models and Computational Methods Program.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.