Machine learning bioactive compound solubilities in supercritical carbon dioxide

The supercritical carbon dioxide (sc-CO₂) is one of promising solvents for supercritical fluid extraction. The solubility of bioactive compounds in the sc-CO₂ is an important factor in designing the extraction process. Obtaining solubility data by experiments requires a large amount of resource and manpower. A model is highly sought if it is simple and can apply to a wide selection of solutes. We develop the Gaussian process regression model to study the relationship among the solute molecular weight, solute melting point, temperature, pressure, sc-CO₂ density, and solubilities of 30 bioactive compounds in more than 1,000 scenarios. The model manifests a high degree of accuracy and stability, contributing to fast low-cost estimations of solubilities of a variety of bioactive compounds.