SusChEM: Machine learning blueprints for greener chelants

1705592 (Keith)。螯合剂最近已被确定为一个关键类别的化学产品，是成熟的绿色设计。据推测，与传统的试错实验相比，确定更好的替代品将需要对化合物空间进行更广泛的探索。在这个项目中，精确的量子化学计算将用于训练最先进的机器学习方法，这些方法将允许预测绿色螯合剂的结构。即将开发的机器学习方法承诺了一种快速预测螯合剂/金属配合物性质的新途径，不仅具有更高的精度，而且比传统的预测量子化学方法（例如Kohn-Sham密度泛函数理论）的计算时间减少了六个数量级。有了这个计算工具，将有可能快速筛选大约10万个假设的螯合剂结构，以预测那些能与不同金属离子强烈结合的结构。该项目还将筛选这些复合体，以确定哪些复合体在合理的时间内具有高降解倾向，哪些复合体具有低毒性。这种新型筛选方法的最佳候选物质将通过实验合成和测试。这将验证基于量子化学的机器学习是否会成为环境可持续性和绿色化学设计的变革性工具，成为传统QSAR模型的更具预测性的补充和/或替代方案。该项目将解决四个基本科学问题。首先，最先进的计算量子化学将用于定量理解哪种螯合剂/金属络合物性质（键能，pKas等）与水溶液中螯合剂的整体稳定性常数最相关。其次，机器学习方法将被开发出来，用于驱动对非传统分子结构的硅搜索，这些分子结构将能够与EDTA等不同的金属离子结合得一样强（或更强）。第三，将使用额外的计算筛选程序来发现这些有前途的螯合剂结构中哪些不太可能是无毒和非持久性的。最后，将通过实验验证哪些通过计算确定的新型螯合剂将被工业合成并在经济上可行以广泛使用。如果成功的话，这项研究将标志着计算机辅助设计更环保的螯合剂的范式转变，这些螯合剂用于洗涤剂、重金属中毒处理、土壤处理中的金属提取、废物修复、从水力压裂场地隔离正常发生的放射性物质，以及水净化。该项目将为引入比广泛使用的QSAR模型具有更大预测能力的新型最先进的计算建模工具奠定重要的基础工作。所有开发的计算机程序以及如何使用这些程序的随附教程将在PI的网站上免费提供。这个项目的教育部分将开发一个电脑游戏，“螯合-它”，这将允许学生量化螯合中不同的基本化学键概念，然后利用这些知识自己设计新的螯合结构。这款电脑游戏将在匹兹堡大学的一个暑期学校项目中测试，对象是代表性不足的10年级学生。电脑游戏教育学生关于化学键、环境可持续性工程和研究的能力将被评估。

1705592 (Keith). Chelating agents have recently been identified as a key category of chemical products that are ripe for greener design. It is hypothesized that identifying better alternatives will require far broader explorations of chemical compound space than what is possible with conventional trial and error experimentation. In this project, accurate quantum chemistry calculations will be used to train state-of-the-art machine learning methods that will allow prediction of structures of greener chelating agents.The machine learning method that will be developed promises a novel route to rapidly predict properties of chelant/metal complexes, not only with higher accuracy but with six orders of magnitude less computational time than conventional predictive quantum chemistry methods (e.g. Kohn-Sham Density Functional Theory). With this computational tool, it will be possible to rapidly screen through about 100,000 hypothetical chelant structures to predict those that would bind strongly to different metal ions. The project will also screen these complexes to see which have high propensities to degrade in reasonable timeframes, and which have low probabilities of being toxic. The top candidates from this novel screening approach will then be experimentally synthesized and tested. This will validate if quantum chemistry-based machine learning would be a transformative tool for environmental sustainability and green chemical design by being a more predictive supplement and/or alternative to conventional QSAR models. Four basic scientific questions will be addressed by the project. First, state-of-the-art computational quantum chemistry will be used to develop a quantitative understanding of which chelant/metal complex properties (bond energies, pKas, etc.) best correlate with overall chelant stability constants in aqueous solutions. Second, machine learning methods will be developed to be used to drive in silico searches for non-traditional molecular structures that would be able to bind as strong (or stronger) to different metal ions as EDTA. Third, additional computational screening procedures will then be used to find which of these promising chelant structures are unlikely to be non-toxic and nonpersistent in nature. Finally, it will be experimentally validated which of the novel chelants identified via computation would be industrially synthesized and economically viable for widespread use. If successful, this research effort would signify a paradigm shift for computer-aided design of greener chelants used in detergents, treatments of heavy metal poisoning, metal extraction for soil treatments, waste remediation, sequestering normally occurring radioactive materials from hydraulic fracturing sites, and water purification. This project will lay important foundational work that is needed to introduce new state-of-the-art computational modeling tools with greater predictive capacity than widely used QSAR models. All developed computer programs and accompanying tutorials for how to use the programs will be made freely available on the website of the PI. The educational component of this project will develop a computer game, "Chelate-it", which will allow students to quantify different fundamental chemical bonding concepts involved in chelation and then use that knowledge to design novel chelant structures on their own. The computer game will be tested in a summer school program at the University of Pittsburgh for underrepresented 10th grade students. The capacity for the computer game to educate students about chemical bonding, environmental sustainability engineering, and research will be assessed.