Machine Learning in Catalysis


An RWTH research group led by Professor Franziska Schoenebeck recently published an article in renowned journal Science.


Chair of Organic Chemistry I Professor Franziska Schoenebeck and her research group published the article Accelerated Dinuclear Palladium Catalyst Identification Through Unsupervised Machine Learning in the journal Science in November 2021.

Molecular research, particularly in catalysis, is predominantly driven by intuition-biased trial-and-error experiments, screening approaches, mechanistic studies, and machine learning methods, such as supervised learning. Although these methods have proven to be powerful tools for discovering new catalysts, their implementation suffers from inherent limitations: Screening approaches are typically limited by the commercial availability of the molecules screened and mechanistic studies are difficult to apply to complex problems. Supervised learning methods require huge amounts of experimental data to be predictive. Some issues have therefore arisen for which none of the tools mentioned above has been able to offer a solution yet.

In their recent Science article, Professor Franziska Schoenebeck's research group reports on a new strategy to solve such problems using unsupervised machine learning. In particular, they investigated which phosphine ligands can stabilize palladium in a dinuclear geometry. To date, only four ligands are known to induce the formation of this particular Pd(I) dimer scaffold. The researchers used an unsupervised machine learning algorithm to discover new ligands that form the Pd(I) dimer and expand the repertoire of chemical reactivity. They identified a total of 21 non-intuitive ligands, which could finally be experimentally verified on representative examples. The new methodology allowed the researchers to consider previously unproduced ligands and verify one of them to induce the formation of the desired Pd(I) dimer. Their study demonstrates the power of machine learning when it comes to navigating and filtering the chemical space and enabling non-intuitive suggestions to accelerate the discovery of new catalysts.

Read the complete article.

Published in:

Science, November 25, 2021:
Vol. 374, Issue 6571, pp. 1134-1140
DOI: 10.1126/science.abj0999