CAS: Developing New Physical Organic Descriptors for Flexible, Large Catalyst Architectures

在化学系化学催化项目的支持下，犹他大学的马修·S·西格曼教授正在开发将机器学习和数据科学与有机反应优化相结合的计算工作流程。Sigman小组的广泛目标是从根本上了解催化剂结构和功能之间的关系。许多现代催化剂结构复杂，自由度高，这对用数值方法描述结构提出了很大的挑战。该计划的目标是开发一个通用的工作流程，使用计算和数据科学工具对柔性催化剂进行参数化，并最终建立描述其反应性能的预测模型。通过这一策略产生的新描述符将在C-H键官能化反应的背景下进行评估，该反应是工业和学术环境中使用的广泛反应的基础。Sigman小组将开发的分子描述符和工作流程也预计将广泛适用于仿生催化和超分子化学。与机器学习相结合的计算描述复杂反应的能力不仅提供了机理上的洞察，而且还能够预测反应结果和催化剂/底物的性能。这些努力的成功将为社区提供一种工具，以简化社会重要化合物的合成。此外，Sigman小组将继续在有机化学/数据科学接口进行各种合作。西格曼和他的团队将与犹他大学的SRI计划一起启动科学研究倡议(SRI)流，将数据科学与化学反应开发相结合。SRI计划是最近建立的一个计划，致力于促进即将到来的大学新生的研究机会。推动拟议活动的主要断言是，数据科学方法的进步，以准确描述分子结构，将提供更精确的解释和复杂反应关联的预测性应用。因此，这项提议的一个中心目标是开发能够描述超分子和现代催化化学中常见的高度灵活和复杂的催化剂结构的计算工作流。待开发的特定参数已被标记为“可接近刚性目标的空间造型”，或智能描述符。这种方法将每个相关催化剂构象的反应位置视为一个“可接近的刚性靶标”，探针分子可以对接到该靶标。然后，通过评估探针分子在催化剂原子冻结的受限构象搜索中可以占据的空间，直接捕获反应位置的空间约束。这一拟议的工作流程具有潜在的深远影响，因为它允许可视化和量化探针分子可能占据的所有可能的笛卡尔空间，给定催化剂口袋的空间限制。Sigman小组计划在涉及卡宾、恶烯和硝烯的C-H官能化反应的背景下评估这些描述符，这些反应基于Davies和Du Bois小组开发的Rh2L4结构。这些都是大型、复杂的建筑。对这些灵活而复杂的催化剂的基本了解将为合成普遍存在的C-H官能化反应提供机理上的洞察，而机器学习和数据科学的使用则有可能使可行的反应结果预测工具成为可能。预计这项工作将对仿生催化和超分子化学具有广泛的、长期的相关性。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。

With support from the Chemical Catalysis program in the Division of Chemistry, Professor Matthew S. Sigman of the University of Utah is developing computational workflows that integrate machine learning and data-science with organic reaction optimization. The Sigman group has the broad goal of fundamentally understanding the relationship between catalyst structure and function. Many modern catalysts are structurally complex and have high degrees of freedom, which presents a significant challenge in representing the structures using numerical measures. The goal of this program is to develop a general workflow using computational and data science tools to parameterize flexible catalysts and ultimately build predictive models that describe their reaction performance. New descriptors generated from this strategy will be evaluated in the context of C‒H bond functionalization reactions, which are at the foundation of a vast range of reactions used industrially and in academic settings. The molecular descriptors and workflows the Sigman group will develop are also envisioned to be broadly applicable to biomimetic catalysis and supramolecular chemistry. The ability to computationally describe complex reactions in tandem with machine learning not only provides mechanistic insight, but it also enables prediction of reaction outcomes and catalyst/substrate performance. Success in these endeavors would provide the community with a tool to streamline the synthesis of important compounds for society. Additionally, the Sigman group will continue diverse collaborations at the organic chemistry/ data science interface. Sigman and his team will begin a Science Research Initiative (SRI) stream on integration of data science with chemical reaction development with the University of Utah’s SRI program, a program recently established and devoted to facilitating research opportunities to incoming college freshman.The overarching assertion driving the proposed activities is that advances in data science methods to accurately describe molecular structure will deliver more precise interpretation and predictive application of complex reaction correlations. Therefore, a central goal of this proposal is to develop computational workflows capable of describing highly flexible and complex catalyst architectures that are commonplace in supramolecular and modern catalytic chemistry. The specific parameters to be developed have been labeled “Spatial Molding for Approachable Rigid Targets”, or SMART descriptors. This approach treats the reactive site of each relevant catalyst conformation as an “approachable rigid target” to which a probe molecule can be docked. The spatial constraints of the reactive site are then directly captured by assessing what space the probe molecule can occupy in a constrained conformational search where the catalyst atoms are frozen. This proposed workflow has the potential for far reaching impact as it allows for visualization and quantification of all possible cartesian space that the probe molecule can occupy, given spatial constraints of the catalyst pocket. The Sigman group plans to evaluate these descriptors in the context of C–H functionalization reactions involving carbenes, oxenes, and nitrenes which are based on the Rh2L4 structure developed by the Davies and Du Bois groups. These are large, complex architectures. A fundamental understanding of these flexible and complex catalysts would provide mechanistic insight to synthetically pervasive C‒H functionalization reactions while the use of machine learning and data science have the potential to enable viable predictive tools for reaction outcomes. It is anticipated that this work will have broad, long term relevance to biomimetic catalysis and supramolecular chemistry.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.