Accelerating catalyst design using reaction-path data mining

催化作用支撑着每年3500亿英镑的全球化学工业，为合成新的抗生素提供了新的途径，从我们呼吸的空气中去除空气污染，或将工业废物转化为有用的产品，如塑料。简而言之，如果没有催化作用，许多我们认为理所当然的产品、药物、燃料和材料将根本不存在。不幸的是，设计具有目标性能的新型催化剂对工业界和学术界来说仍然是一个巨大的挑战。关键原因是复杂性；当代的多相和纳米颗粒催化剂可以表现出令人难以置信的反应位点和途径范围，催化剂活性可以依赖于（通常以不明确的方式）广泛的特征，如结构、组成、支撑相互作用、温度、压力、反应物组分和副产物中毒。这种巨大的化学复杂性直接阻碍了世界各地采用传统的试错合成方法来设计催化剂。但是，如果我们可以教计算机自动设计新的、更好的催化物种呢？这将对学术界和工业界的催化研究产生变革性影响；使用计算机准确预测反应的最佳催化剂将减少在试错合成中浪费的时间，加速催化剂的发现并提高可持续性。然而，迄今为止，催化剂的自动化计算设计已经被证明是难以捉摸的；同样，影响实验催化剂设计的化学复杂性问题也同样阻碍了计算方法。这个项目旨在改变这种情况，推动我们朝着计算催化剂设计的“黑匣子”策略发展。具体来说，我们将开始使用路径约束分子动力学（PCMD）来解决这一挑战，PCMD是PI最近开发的一种新的计算方法。PCMD是一种连接驱动的采样策略，可以快速生成连接大量不同化学物质的反应路径；结合反应速率的量子化学计算和动力学模型，PCMD支撑了一种分层策略，可以预测由于催化剂特性变化而引起的速率定律、选择性和产物产量的趋势。据我们所知，PCMD是第一个能够预测复杂催化转化（如烯烃氢甲酰化）的紧急机制和速率规律的自动化“黑匣子”策略。在PCMD的第一次工业应用中，我们将寻求对纳米颗粒和非均相催化系统的反应化学产生新的见解，以控制废气排放。我们将与世界领先的排放控制技术公司庄信万丰合作，利用PCMD制定纳米颗粒和非均相催化剂上关键废气反应的反应机理、热力学和动力学的“路线图”，特别是金属纳米颗粒和cu促进沸石上的一氧化碳氧化和氮氧化物还原。此外，与华威数据科学研究所和艾伦图灵研究所建立新的合作关系，我们将对PCMD生成的（潜在巨大的）反应路径数据集应用“大数据”统计分析；这导致了反应路径数据挖掘（RDM）的新概念，它将反应路径数据集转化为催化剂功能的切实见解和描述符。总的来说，我们的PCMD/RDM策略代表了计算催化的新方向；通过大大加快这一战略的发展和应用，该项目将成为我们最终长期目标的关键里程碑，即新的催化剂，分子和其他功能化学系统的“黑匣子”计算设计。

Catalysis underpins the £3,500B/year global chemical industry, enabling new routes to synthesising new antibiotics, removing air pollution from the air we breathe, or turning industrial waste into useful products such as plastics. In short, without catalysis, many of the products, drugs, fuels and materials we take for granted would simply not exist.Unfortunately, the design of new catalysts with targeted properties remains an enormous challenge to industry and academia. The key reason is complexity; contemporary heterogeneous and nanoparticle catalysts can exhibit a mind-boggling range of reaction sites and pathways, and catalyst activity can depend (often in an ill-defined manner) on a wide range of features such as structure, composition, support interactions, temperature, pressure, reactant phase constituents, and by-product poisoning. This enormous chemical complexity is a direct barrier to the traditional trial-and-error synthetic approaches to catalyst design used the world over.But, what if we could teach computers to automatically design new, better, catalytic species instead? This would have a transformative impact on catalysis research, both in academia and industry; using computers to accurately predict the optimal catalyst for a reaction would cut down time wasted in trial-and-error synthesis, accelerate catalyst discovery and improve sustainability. However, automated computational design of catalysts has proven elusive to date; again, the same issue of chemical complexity which dogs experimental catalyst design similarly hinders computational methods.This project aims to change this situation, pushing us towards development of a "black box" strategy for computational catalyst design. Specifically, we will begin to address this challenge using path-constrained molecular dynamics (PCMD), a new computational approach developed recently by the PI. PCMD is a connectivity-driven sampling strategy which enables rapid generation of reaction paths connecting large numbers of different chemical species; combined with quantum-chemical calculations of reaction rates and kinetic modelling, PCMD underpins a hierarchical strategy which can predict trends in rate laws, selectivities and product yields arising as a result of changes to catalyst features. To the best of our knowledge, PCMD was the first automated "black box" strategy shown capable of predicting the emergent mechanism and rate law of complex catalytic transformations such as alkene hydroformylation. In the first industrial application of PCMD, we will seek to generate new insights into the reactive chemistry of nanoparticle and heterogeneous catalytic systems for exhaust emissions control. In collaboration with Johnson Matthey, a world-leader in emissions control technologies, we will use PCMD to develop a 'roadmap' of reaction mechanisms, thermodynamics and kinetics of key exhaust gas reactions on nanoparticle and heterogeneous catalysts, specifically carbon monoxide oxidation and nitrogen oxide reduction on metallic nanoparticles and in Cu-promoted zeolites. In addition, building new collaborations with Warwick Data Science Institute and The Alan Turing Institute, we will apply 'big data' statistical analyses of the (potentially enormous) reaction-path datasets generation by PCMD; this leads to the new concept of reaction-path data mining (RDM), which will transform reaction-path datasets into tangible insights and descriptors of catalyst function. Overall, our PCMD/RDM strategy represents a new direction for computational catalysis; by dramatically accelerating the development and application of this strategy, this project will be a critical milestone towards our ultimate long-term goal, namely the "black box" computational design of new catalysts, molecular and other functional chemical systems.