ESA Toolbox - Vanadium-Catalyzed Epoxidation Reaction Database

We present the full database of the New Journal of Chemistry article "Navigating epoxidation complexity: building a data science toolbox to design vanadium catalysts".

This database stems from the methodology built for a Machine Learning (ML) ready database of vanadium-catalysed epoxidation of small alcohols and alkenes (ESA). It amasses 273 distinct reactions, featuring five different vanadium-catalyst scaffolds: vanadyl(iv) sulphate – [VOSO4], vanadyl acetylacetonate-salen – [VO(salen)], vanadyl isopropoxide – [VO(OiPr)3], vanadyl acetylacetonate – [VO(acac)2], and vanadyl dichloride-salen – [VCl2(salen)]. 

Each reaction component – catalysts, ligands, substrates, oxidants, and solvents – was meticulously represented in a 3D structure using MarvinSketch. This was complemented by additional experimental data extracted from the literature, encompassing enantiomeric excess, solvent and oxidant type. Source code and 3D structures are presented in the article's repository.

This dataset integrates a comprehensive set of descriptors, meticulously chosen to capture the nuances of epoxidation reactions, including experimental reaction descriptors and solvent characteristics. Molecular descriptors were calculated by the open-source RDKit software, using each SMILES string (generated from the 3D structure) as input, and were assorted into different chemical groups: volume surface area, electronic and structural descriptors. The collected data underwent rigorous processing, including the normalization of units, handling of missing values, and standardisation of formats. The final dataset comprised more than 90 000 data entries, while the targets were specific reaction outcomes such as yield and enantiomeric excess.

Utilising data-driven analysis, we identified key reaction yield trends through chemical descriptors, providing insights for catalyst design and reaction optimisation.