Experimental Catalyst Design Aided by Machine Learning

In the past, catalyst design has been highly characterized by trial and error based on previous experience and knowledge of catalysis.¹ Due to the number of variables involved in catalyst performance, it is difficult to manually optimize highly active catalysts. In the past few years, there has been a movement toward the use of machine learning as a tool in experimental catalyst design. There have been numerous studies published using different machine learning methods to aid catalyst design in different ways, including predicting optimal experimental conditions for high catalyst performance and directly designing catalysts to be used in specific reactions.^2-3 Recently, Nguyen et al. proposed a high-throughput screening method to produce large quantities of data, which could then be used by machine learning for catalyst design. They used this instrument to generate data by testing the performance of different catalysts with different experimental conditions for the catalytic oxidative coupling of methane. They then used this data to train machine learning methods to predict the optimal experimental conditions for high catalyst performance.² A second study used this same data set to train a deep learning method to directly design high performance catalysts for this reaction.³ The application of machine learning algorithms to optimize catalytic performance has great potential to accelerate the predictive discovery of novel catalyst formulations. 

1. Z. Li, S. Wang, H. Xin Nature Catalysis 1, 641 (2018)
2. T.N. Nguyen, T.T.P. Nhat, K. Takimoto, A. Thakur, S. Nishimura, J. Ohyama, I. Miyazato, L. Takahashi, J. Fujima, K. Takahashi, T. Taniike ACS Catal. 10, 921 (2020)
3.  K. Sugiyama, T.N. Nguyen, S. Nakanowatari, I. Miyazato, T. Taniike, K. Takahashi Chem. Cat. Chem. 13, 952 (2021)