CDS&E: D3SC: The Dark Reaction Project: A machine-learning approach to exploring structural diversity in solid state synthesis

该奖项由材料研究部、化学部和先进网络基础设施办公室资助。该奖项支持使用以数据为中心的方法来预测具有所需性质的金属氧化物化合物的研究和教育。有机模板金属氧化物具有极大程度的结构多样性和组成灵活性。这使得化学家可以调整这些化合物的结构，性质和对称性，以优化其在特定应用中的性能，包括催化，分子筛分，气体吸附和非线性光学。然而，新化合物通常是通过试错过程产生的，在固态化学中，创造具有特定结构的新化合物是一个巨大的挑战。该项目将为计算机开发人工智能技术，称为机器学习技术，可用于预测化学反应的条件，从而增加结构多样性并导致特定的结构特征。该项目还将开发机器学习技术，生成关于形成机制的人类可读解释，这将在实验室进行测试。该项目的主要影响将是减少发现具有特定结构特征的新材料的时间和成本，这反过来有助于将新材料更快地应用于市场。这个项目是合成化学家、计算化学家和计算机科学家之间合作的一个例子，作为一个模型，它可以直接转移到广泛的学科和研究途径。本科生的研究机会和课程发展将贯穿整个项目，从而为科学劳动力做出贡献。该奖项的资金来自材料研究部、化学部和先进网络基础设施办公室。该奖项支持使用以数据为中心的方法来预测具有所需性质的金属氧化物化合物的研究和教育。水热合成技术被广泛用于制备具有广泛功能特性和应用的新型金属氧化物材料。该项目将通过开发软件基础设施将x射线衍射实验结果与单个反应联系起来，从这些数据中提取结构结果描述符，然后确定从反应描述数据中预测这些结构结果的程度，从而推进该领域的发展。这将通过开发几何性质、非共价相互作用性质和电子密度性质的结构结果描述符来实现，然后建立将这些结果与反应条件相关联的机器学习模型，最后通过实验测试这些预测的质量。主动学习和可审计和可解释的模型将被纳入工作流程，以帮助合成化学家以互动的方式选择更好（更有洞察力/新颖）的反应。

NONTECHNICAL SUMMARYThis award receives funds from the Division of Materials Research, the Chemistry Division and the Office of Advanced Cyberinfrastructure. This award supports research and education that uses data-centric methods to enable the prediction of metal oxide compounds with desired properties. Organically-templated metal oxides have a tremendous degree of structural diversity and compositional flexibility. This allows chemists to tune the structures, properties, and symmetries of these compounds to optimize their performance in specific applications that include catalysis, molecular sieving, gas adsorption, and nonlinear optics. However, new compounds are typically created by a trial-and-error procedure, and creating novel compounds with specific structures is a grand challenge in solid state chemistry. This project will develop artificial intelligence techniques for computers called machine learning techniques that can be used to predict the conditions for chemical reactions that will increase structural diversity and lead to specific structural features. This project will also develop machine learning techniques that generate human-readable explanations about the formation mechanism, which will be tested in the laboratory. The primary impact of this project will be to decrease the amount of time and to lower the cost of discovering new materials with specific structural features, which in turn help bring new materials for applications to market more quickly. This project is an example of a collaboration among synthetic chemists, computational chemists, and computer scientists and as a model it may be directly transferred to a wide range of disciplines and avenues of investigation. Undergraduate student research opportunities and curricular developments will be involved throughout the project, thus contributing to the scientific workforce. TECHNICAL SUMMARY This award receives funds from the Division of Materials Research, the Chemistry Division and the Office of Advanced Cyberinfrastructure. This award supports research and education that uses data-centric methods to enable the prediction of metal oxide compounds with desired properties. Hydrothermal synthesis is widely used to create new metal oxide materials with a wide range of functional properties and applications. This project will advance the field by developing software infrastructure for associating the results of X-ray diffraction experiments with individual reactions, extracting structural outcome descriptors from this data, and then determining the extent to which these structural outcomes can be predicted from reaction description data. This will be achieved by developing structural outcome descriptors for geometric properties, non-covalent interaction properties, and electron-density properties, then building machine learning models that correlate these outcomes to reaction conditions, and finally testing the quality of these predictions experimentally. Active learning and auditable and interpretable models will be incorporated into the workflows to help synthetic chemists select better (more insightful/novel) reactions in an interactive fashion.