TasteMolNet: A machine learning-driven platform for sweet, bitter, and tasteless compounds prediction in food chemistry

The identification of sweet and bitter compounds is essential for improving the sensory quality of foods, yet traditional analytical methods remain time-consuming and costly. To overcome this limitation, we developed a machine learning framework for high-throughput taste prediction of sweet, bitter, and tasteless compounds. Our optimized model achieved an F1 score of 88.0 %, and its reliability was validated through electronic tongue analysis and molecular docking. Furthermore, we implemented this model into an online platform named TasteMolNet ( http://www.bstchem.fun/ ), which serves as a practical tool to accelerate sweetener discovery and bitterness modulation in food research. • Constructed a curated dataset of sweet, bitter, and tasteless compounds. • Identified XGBoost-MACCS as the most accurate taste-prediction model. • Integrated SHAP, e-tongue testing, and docking for model validation. • Developed TasteMolNet for rapid prediction of food-compound taste modalities.