Conventional extraction of valuable plant compounds often relies on hazardous volatile organic solvents (VOCs), posing environmental and health risks. This study explores a sustainable alternative using Natural Deep Eutectic Solvents (NADES) designed in-silico through the Conductor-like Screening Model for Realistic Solvents methodology (COSMO-RS) for efficient extraction of target plant metabolites. A library of NADES with varying compositions was designed using COSMO-RS to predict their physicochemical properties and affinity for target natural compounds, selecting the most promising candidates in terms of versatility, cost-effectiveness and biocompatibility. To complete the study, a predictive Artificial Intelligence based method (Decision Trees) was developed for reverse design of NADES for target bioactive compounds from energetic and structural molecular descriptors. From a compendium of 58 plant metabolites of interest and 66 natural compounds as NADES components, nearly 3000 solubility in silico tests were conducted and a total of 12 NADES were selected. Three solubility models were created, and a clear dependance of the target compound properties was observed.
