Abstract: Natural deep eutectic solvents (NADES) show great promise as media for enzymatic reactions in areas where (bio)compatibility with natural or medicinal products is a must. While in theory they can be tailored to the intended reaction to ensure optimized yields, the knowledge to date is predominantly empirical, with some mechanistic reports providing a fragmented view at best. Therefore, it is not easy to explain experimental observations, let alone make predictions. The aim of this study was to develop a structured, holistic understanding of the effects of NADES media on enzymatic reactions, distinguishing between effects on solubility, solvation, viscosity, inhibition and denaturation. Experimental and computational chemistry methods were combined to separately study the interactions between enzyme, substrate, and NADES as reaction media. The initial enzyme activity and the final conversion of vinyl laurate transesterification by immobilized Candida antarctica lipase were studied experimentally. The direct effect of NADES on the same enzyme was modeled by molecular dynamics simulation. The effect of solubility was studied by both experimental and computational methods. To predict the solubility and viscosity of NADES, data-driven models were developed by combining group contribution and machine learning methods, based on the accumulated experimental knowledge on NADES found in the literature. Finally, the composed relationships and prediction models were applied to the practical example of deacetylation of mannosylerythritol lipids (MELs). The experimental findings show that the chosen NADES system has a significant effect on both the apparent initial activity and the final conversion. However, in the simulations, the enzyme retains its original structure; moreover, NADES has an additional stabilizing effect on the enzyme. In addition, changes in the molar ratio of the compounds in NADES do not show a significant effect on the stability of the enzyme. These results indicate that the main effect of NADES on the reaction is mainly related to the substrate-solvent interactions (solvation energy) and the viscosity of the system. On the other hand, the experimental results only confirmed the significance of solvation, viscosity did not show a clear correlation with the studied reaction parameters. The machine learning models built for solubility and viscosity gave quantitative predictions of these properties. The accumulated knowledge was used to optimize the yield in the deacetylation reaction of MELs. The combination of these methods provides fundamental knowledge about the effect of NADES on biocatalysis, but the results are also applicable to other uses of NADES.