Advances in machine learning screening of food bioactive compounds

Food contains abundant bioactive compounds that play a crucial role in human health. However, conventional methods for screening food bioactive compounds (FBCs) are expensive, time-consuming, and labor-intensive. Machine learning (ML) techniques, offer an efficient and cost-effective means to screen potential bioactive compounds, but their application and research in the food field are limited. In order to facilitate the effective screening of FBCs and to provide researchers with valuable insights, this paper presents the process of constructing ML models, including data preparation, molecular representation, ML algorithms selection, and evaluation methods, and highlights the progress of ML in screening FBCs with different bioactivities in recent years. Furthermore, this paper puts forward the primary limitations and challenges and suggests future directions. The research on using ML for screening FBCs has made some progress, especially for peptides with antioxidant and antihypertensive activities, as well as non-peptidic compounds with hypoglycemic activity. Different limitations and challenges exist in establishing ML models for different bioactivities. To ensure high-quality ML models, each step of the modeling process needs to be carefully considered based on the actual situation. In the future, to further promote the application of ML in the food field, it is necessary to establish comprehensive databases of FBCs, improve the quality and quantity of datasets, enhance the interpretability of deep learning; integrate ML with other techniques; combine classification and regression models; comprehensively evaluate compounds from multiple aspects; and develop models that can predict multiple bioactivities.