Machine learning insights into important catalytic reactions
03 09 2026
Category: Main page, Research highlights
The conversion of plant oils into platform chemicals offers a sustainable alternative to petrochemical manufacturing. A key reaction in this transition is ethenolysis, where fatty esters like ethyl oleate are cleaved with ethylene and ruthenium catalysts to yield terminal olefins. The main hurdle remains selectivity. Catalysts frequently trigger a competing self-metathesis route, consuming raw materials to generate heavy, low-value internal dimers instead of the desired products.
In their latest publication on this topic, the team of the Chemical and Biological Systems Simulation Laboratory uncovered, via computational chemsitry appoaches, that selectivity i ethenolysis is not dictated by kinetic activation barriers, but by thermodynamics during catalyst recovery. The overall reaction energy of the productive cycle governs the experimental outcome: catalysts that make catalyst regeneration more exergonic systematically outperform the rest. Rigid spirocyclic carbenes (spiro-CAACs), in particular, exhibit a strong thermodynamic driving force for catalyst recovery compared to traditional N-heterocyclic carbenes, suppressing the undesired pathway.
To scale the resutlts to larger catalyst libraries, they built a binary classification framework using Random Forest algorithms to filter viable candidates from non-selective ones. The machine learning models demonstrated that catalytic selectivity cannot be screened through single electronic descriptors like metal electrophilicity alone. Instead, selectivity emerges from multidimensional interplay between steric congestion and d-orbital populations. This hybrid computational chemsitry – machin learning framework acts as a high-throughput computational filter, discarding unselective architectures in silico before entering the laboratory.
Daniel E. Trujillo-González, Kamil Szychta, Albert Poater, Juan V. Alegre-Requena, Bartosz Trzaskowski, J. Pablo Martínez, DFT and machine learning insights into productive versus self-metathesis selectivity in ruthenium-catalyzed ethenolysis, Journal of Catalysis, https://doi.org/10.1016/j.jcat.2026.117142.