Machine Learning in Catalysis

Computational chemistry has become an increasingly common part of catalysis research. More recently, data-based methods such as machine learning have been suggested as a means to speed up discovery. This Focus issue features a collection of content dedicated to machine learning as pertaining to its potential impact on the field of catalysis.

Data science and machine learning have the potential to accelerate the discovery of effective catalysts; however, these approaches are currently held back by the issue of negative results. This Comment highlights the value of negative data by assessing the bottlenecks in data-driven catalysis research and presents a vision for a way forwards.

Reaction networks provide complete mechanistic understanding of catalytic processes, although they can be highly complex and thus very challenging to obtain. This Perspective discusses the use of machine learning for the exploration of reaction networks in heterogeneous catalysis.

Computational chemistry has the potential to aid in the design of heterogeneous catalysts; however, there is currently a large gap between the complexity of real systems and what can be readily computed at scale. This Review discusses the ways in which machine learning can assist in closing this gap to facilitate rapid advances in catalyst discovery.

Retrobiosynthesis aims to create novel biosynthetic pathways for the beneficial production of molecules of interest. This Review outlines how machine learning can help to advance retrobiosynthesis by improving retrosynthesis planning, enzyme identification and selection, and the engineering of enzymes and pathways.

Comprehensive information on enzyme catalytic rates is essential to understand the metabolism of cells, but only a small fraction has been determined experimentally. Now, a deep learning model is developed to predict k_cat values of metabolic enzymes on a large scale using substrate SMILES and protein sequence information.

Most applications of machine learning in catalysis use black-box models to predict physical properties, but extracting meaningful physical insights from them is challenging. This Perspective discusses machine learning approaches for heterogeneous catalysis and classifies them in terms of their interpretability.

Proton exchange membrane water electrolysers require the development of active, stable and cost-effective catalysts for water oxidation. Now, a Ru/α-MnO₂ catalyst with in-situ-formed arrays of Ru atoms is presented for acidic water oxidation, which follows the oxide path mechanism and achieves enhanced activity and stability.

Metal oxide alloys are important industrial catalysts, but their structure–activity relationships are poorly understood. Now, a study encompassing a combination of computational tools and machine learning approaches sheds light on the activity and selectivity of zinc–chromium oxides during syngas conversion.

The design of new catalysts for electrochemical energy storage is of utmost importance. Here, an automated computational screening method is used to identify over 100 intermetallic surfaces as efficient electrocatalysts for CO₂ reduction and H₂ evolution.

Predicting metal–support combinations that can afford stable single-atom catalysts remains a complex problem. Now, a computational method is reported that can be used to screen interaction strengths between metals and supports and identify those pairs that generate strongly adsorbed single-atom catalysts.

Catalysis is a complex, multidimensional and multiscale field of research. Machine learning is helping to build better models, understand catalysis research and generate new knowledge about catalysis.