Identification of natural food-derived emulsifiers using QSAR and machine learning: Application in dairy emulsions

Emulsifiers maintain the stability of emulsions, and milk protein-formed emulsions are unstable. Hence, efficient ways to screen food-derived compounds need to be identified. This study combined molecular descriptors with machine learning algorithms to construct quantitative structure-activity relationship models for screening emulsifying compounds. The candidate compound was added to different milk protein emulsions to assess their stability. The results indicated that the model built with the molecular operating environment descriptor combined with the random forest algorithm had the optimal prediction ability (area under the curve: 0.9947; accuracy: 93.33 %). Glycyrrhizin has excellent emulsifying ability, and the emulsion droplets prepared by combining glycyrrhizin and milk protein have a smaller particle size and potential (102.64 nm and − 36.64 mV, respectively), better physical stability, improved foaming properties by 39.59 %, and higher viscosity. Glycyrrhizin significantly improved the physical stability of milk protein emulsions, providing new perspectives for the food industry.