Using Artificial Intelligence To Forecast Water Oxidation Catalysts

Abstract

Artificial intelligence and various types of machine learning are of increasing interest not only in the natural sciences but also in a wide range of applied and engineering sciences. In this study, we rethink the view on combinatorial heterogeneous catalysis and combine machine learning methods with combinatorial approaches in electrocatalysis. Several machine learning methods were used to forecast water oxidation catalysts on the basis of literature published data sets and data from our own work. The machine learning models exhibit a decent prediction precision based on the data sets available and confirm that even simple models are suitable for good forecasts.

Artificial‐Intelligence‐Driven Organic Synthesis—En Route towards Autonomous Synthesis?

“…for object and speech recognition, fundamentally expanded the ability of neuronal networks in data analysis and pattern identification and are today one of the key factors in machine learning algorithms that have led to fascinating achievements, for example, the AlphaGo system, speech recognition and output, and self‐driving cars. The potential for AI‐driven chemistry was recently described in applications ranging from reaction methodology to drug discovery, catalyst design, and the directed evolution of enzymes . These initial reports, which all constitute landmark papers on their own, already outline and foreshadow the disruptive nature of machine learning and its impact on modern chemical sciences.…”

“…With the emergence of artificial intelligence (AI)-driven technology,t he challenge of generalized predictions of reaction outcomes and thus streamlined reaction optimization has gained significant interest in organic synthesis.AIis considered one of the disruptive innovations with the potential to revolutionize everyday life in as imilar fashion and scale as the innovation of the steam engine or more lately internet and communications technology. [2] Deep learning algorithms,introduced in 2012 by Hinton et al for object and speech recognition, [3] fundamentally expanded the ability of neuronal networks in data analysis and pattern identification and are today one of the key factors in machine learning algorithms that have led to fascinating achievements,f or example,t he AlphaGo system, speech recognition and output, and self-driving cars.T he potential for AI-driven chemistry was recently described in applications ranging from reaction methodology [4] to drug discovery, [5] catalyst design, [6] and the directed evolution of enzymes. [7] These initial reports,w hich all constitute landmark papers on their own, already outline and foreshadow the disruptive nature of machine learning and its impact on modern chemical sciences.…”

Universality in Oxygen Evolution Electrocatalysis: High‐Throughput Screening and a Priori Determination of the Rate‐Determining Reaction Step

“…[18,24] Regrettably, the OER kinetics is quite slow so that even for active catalysts elevated overpotentials of η OER = (U-U 0 OER ) > 350 mV are required to sustain a satisfying current density in the order of 10 mA/cm 2 . [27,28] Therefore, an applied overpotential of 400 mV is chosen in the analysis, [18,24] which coincides with typical overpotentials in electrolyzers. [29] Thus, the ESSI-ΔG 2 activity map is compiled at η OER = 0.40 V as a function of the scaling relation's intercept, choosing a step-size of 0.2 eV (i. e., offsets of 2.6 eV, 2.8 eV, 3.0 eV, 3.2 eV, 3.4 eV, and 3.6 eV are inspected).…”

A Primer about Machine Learning in Catalysis – A Tutorial with Code

“…But there are also studies from other fields of Chemistry and Catalysis. With a suitable dataset water oxidation catalysts are predictable with ML [8] and even for approaches from Organic Synthesis there are approaches to make ML based predictions. [9] But especially for people newly approaching this field with a view from their respective discipline Data Science (DS) and Machine Learning sometimes seem to be some kind of arcane art.…”