CAREER: Merging Graph Theory and Automation for Chemical Synthesis

在化学系化学合成项目的支持下，密歇根大学的Timothy Cernak正在建立一个将化学合成与数据科学相结合的项目。这一点很重要，因为化学反应是发明未来的药物、聚合物、农用化学品和储能材料所必需的;然而，目前很难准确预测这些过程的结果。需要公开可用的化学反应数据，以及可以改进预测反应模型的算法和理论，即使在高水平的目标复杂性下也可以在分子合成中应用。 该项目利用最先进的自动化和机器人系统，使用高通量实验（HTE）策略执行数千个化学反应。然后，这些实验的数据将与机器学习、人工智能和其他形式的数据科学合作，旨在建立强大的化学反应性预测模型。为了促进更广泛的社区对机器学习模型的推进，数千个机器可读格式的化学反应数据点将被发布到公共领域。Cernak博士及其研究小组领导的一系列教育活动扩大了该奖项的广泛影响，这些活动侧重于通过编码练习和使用免费软件向本科生和研究生介绍化学合成和数据科学的界面。幼儿也被纳入教育和外展计划，并提供适合其年龄的活动，旨在通过玩井板颜色混合玩具，为幼儿提供自己的高通量实验体验。该奖项支持的研究分为三个研究主题，都在化学合成和数据科学的界面上，如下：（1）识别具有高合成战略价值潜力的新转化，（2）实验实现新型反应类型，（3）基于人工智能的逆合成分析和后续实验路线验证。在第一种情况下，可以想象的反应计算研究使用图论为基础的方法来照亮未来发展的关键转变。例如，邻接矩阵映射技术将应用于萜烯、生物碱和药物的全合成，以揭示尚不存在但仍可能成为合成化学工具箱中有影响力的补充的化学转化类型。在第二个推力中，理论上的声音反应将使用高通量实验和机器学习方法的合并进行实验开发。应用集中在从胺和羧酸结构单元形成碳-氧键。这些反应方法的应用将得到小型化高通量实验整体技术进步的补充，包括硬件和软件层面，以及机器学习算法的开发。最后一个主题侧重于复杂目标分子的多步合成，如药物和代表性生物碱（例如，stemoamide和gelsemine），其中通过逆合成算法将化学反应序列缝合在一起。总的来说，这项研究预计将在化学合成和数据科学的接口推进基础知识和教育。这个奖项反映了NSF的法定使命，并已被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

With the support of the Chemical Synthesis Program in the Division of Chemistry, Timothy Cernak of the University of Michigan is building a program that merges chemical synthesis with data science. This is important because chemical reactions are required to invent the medicines, polymers, agrochemicals, and energy storage materials of the future; however, at the present time, it is difficult to predict the outcome of such processes with accuracy. Publicly available chemical reaction data are needed, alongside algorithms and theories that can improve predictive reaction models toward applications in molecular synthesis even at high levels of target complexity. This project leverages state-of-the-art automated and robotic systems to perform thousands of chemical reactions using the strategy of high-throughput experimentation (HTE). Data from these experiments will then be partnered with machine learning, artificial intelligence, and other forms of data science with the aim of building robust predictive models of chemical reactivity. To facilitate the advancement of machine learning models by the wider community, thousands of chemical reaction data points in a machine-readable format will be released to the public domain. The broader impacts of the award are extended by a range of educational activities being spearheaded by Dr. Cernak and his research group that focus on introducing undergraduate and graduate students to the interface of chemical synthesis and data science through coding exercises and the use of freely available software. Young children are also included in the educational and outreach plans with age-appropriate activities designed to provide kindergartners with their own experiences of high-throughput experimentation through play with a well plate color-mixing toy.The research supported by this award is divided across three investigational themes, all at the interface of chemical synthesis and data science, as follows: (1) identification of new transformations with the potential for high strategic value in synthesis, (2) experimental realization of novel reaction types, and (3) AI-based retrosynthetic analysis and subsequent experimental route validation. In the first case, conceivable reactions are computationally studied using graph theory-based methods to illuminate key transformations for future development. For instance, adjacency matrix mapping techniques will be applied to total syntheses of terpenes, alkaloids, and pharmaceuticals to reveal chemical transformation types that do not yet exist but that could nonetheless become impactful additions to the synthetic chemistry toolbox. In the second thrust, theoretically sound reactions will be experimentally developed using a merger of high-throughput experimentation and machine learning methods. Applications center on the formation of carbon–oxygen bonds from amine and carboxylic acid building blocks. These reaction method applications are to be complemented by the advancement of overall techniques for miniaturized high-throughput experimentation, both at the hardware and software level, as well as in the development of machine learning algorithms. The final theme focuses on multistep synthesis of complex target molecules such as pharmaceuticals and representative alkaloids (e.g., stemoamide and gelsemine), where sequences of chemical reactions are stitched together via retrosynthetic algorithms. Collectively, this research is anticipated to advance basic knowledge and education at the interface of chemical synthesis and data science.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.