Data Science Guided Organic Reaction Development

项目总结 我们计划的总体目标是定义由数据科学驱动的通用工作流 结合物理有机法则，并可直接部署在反应中 优化过程。成功开发这样的工作流将产生三个关键影响 关于化工合成企业：1)显著简化了经验、成本高昂的流程 反应优化，2)算法将被应用于预测新的底物，催化剂， 并且试剂(以及反应条件)在感兴趣的反应中执行AS 这类推断的直觉很差。能够定量地了解 反应的概括性将迅速加速化学中新方法的吸收 综合。以及3)由于这里描述的数据驱动工具利用物理有机方法来 从数学上描述分子，从经验数据得到的关联可以 被解释为提供对催化剂/底物如何相互作用的机械性见解。这 为人们将知识“转移”到新的反应和开发一般知识提供基础 催化剂设计原则。我们计划继续向社会提供一个令人信服的理由 改变反应发展的文化，从经验优化和观察转变为 一个有洞察力的、高效的、高质量的数据产生过程。这项工作将会完成 在不对称催化的背景下，重点关注以下问题：我们能否开发工具 预测培训中未出现的全新示例的反应结果 初始关联所需的数据集，同时具有可解释/可解释 统计模型？这将通过探索各种对映体选择性过程来实现。 由多种催化剂催化，并使用现代技术询问工艺 计算化学和统计学方法。我们将通过以下方式验证这些新方法 探索是否可以使用数据挖掘和新的数据收集来建立与结构的关联 分子的特性用于预测全新的例子。在这一点上，我们将要求 关于催化剂动力学如何与非共价相互作用耦合的基本问题 对催化剂性能的影响以及如何为新的催化剂设计汇编这些信息 战略。最终，我们计划为社区提供一个平台和途径，以促进 反应乐观主义采用了易于应用的数据科学方法。

PROJECT SUMMARY The overarching objective of our program is to define general data science driven workflows that incorporate physical organic precepts and can be deployed directly within the reaction optimization process. Successfully developing such a workflow would have three key impacts on the chemical synthesis enterprise: 1) significantly streamline the empirical, costly process of reaction optimization, 2) algorithms would be applied to predict how new substrates, catalysts, and reagents (as well as reaction conditions) perform in the reaction of interest as extrapolations of this sort are poorly intuited. The ability to know quantitatively the generalizability of a reaction will rapidly accelerate the uptake of new methods in chem ical synthesis. And 3) as the data driven tools described herein utilize physical organic methods to describe molecules mathematically, the resulting correlations derived from empirical data can be interpreted to provide mechanistic insights into how catalysts/substrates interact. This provides one with the foundation to “transfer” knowledge to new reactions and develop general catalyst design principles. We plan to continue to deliver to the community a compelling reason to change the culture of reaction development from empirical optimization and observations to an insightful, efficient, and high quality data producing process. This work will be accomplished in the context of asymmetric catalysis and focus on the following question: can we develop tools to predict reaction outcomes for completely new examples not represented within the training dataset required for the initial correlation, while simultaneously having interpretable/explainable statistical models? This will be accomplished by exploring various enantioselective processes catalyzed by a multitude of catalysts and interrogating the processes using modern computational chemistry and statistical methods. We will validate these new approaches by exploring if data-mining and new data collection can be used to build correlations with structural features of molecules for the prediction of altogether new examples. Within this we will ask fundamental questions about how catalyst dynamics coupled with non-covalent interactions impact catalyst performance and how to compile this information for new catalyst design strategies. Ultimately, we plan to deliver to the community a platform and pathway to facilitate reaction optimism holistically using easy to apply data science methods.