Molecular and Computational Basis of Taste Perception: A Review toward the "Digital Language of Taste"

Taste is a multisensory individualized perception. The variability of genetic and physiological factors and the lack of standardized or detailed experimental protocols for research on taste perception create many methodological difficulties in studies. Recent advances in computational chemistry, molecular modeling, and machine learning (ML) provide new avenues to model taste mechanisms, predict taste profiles, and design novel taste compounds, particularly focusing on G-protein-coupled receptors and ion channels.

This review synthesizes current computational approaches to taste research, including molecular docking, molecular dynamics (MD), and ML. It highlights taste receptors' structural and functional modeling across all primary modalitiessweet, bitter, umami, salty, and sour. Specific focus is given to the challenges of modeling salt and sour taste, integrating MD with receptor-ligand interactions, and applying ML algorithms to predict taste characteristics from molecular descriptors. Recent developments in artificial intelligence (AI) models, such as deep learning and transformer architectures, are improving accuracy but still raise questions regarding the interpretability and generalizability of the data.

Despite advances in taste perception, significant limitations remain. One of the primary factors is incomplete structural data on taste receptors and problems with modeling the long-time scales of receptor activation. This leads to inadequate models for multisensory integration. Future efforts should prioritize high-resolution receptor modeling, hybrid computational-experimental validation, and the expansion of AI applications to generate receptor-specific compounds. Bridging computational predictions with human subjective experience will be key to advancing digital taste perception.