AI-driven prediction of bitterness and sweetness and analysis of receptor interactions
Understanding the molecular mechanisms governing sweetness and bitterness is essential for identifying desirable taste characteristics in natural and synthetic compounds. In this study, we developed graph neural network (GNN)-based artificial intelligence (AI) models to predict bitterness and sweetn...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-01-01
|
| Series: | Current Research in Food Science |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2665927125001212 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849744300712132608 |
|---|---|
| author | Hiroaki Iwata |
| author_facet | Hiroaki Iwata |
| author_sort | Hiroaki Iwata |
| collection | DOAJ |
| description | Understanding the molecular mechanisms governing sweetness and bitterness is essential for identifying desirable taste characteristics in natural and synthetic compounds. In this study, we developed graph neural network (GNN)-based artificial intelligence (AI) models to predict bitterness and sweetness based on chemical structure. GNNs utilize deep learning to capture relationships among molecular components within a graph, extracting latent molecular vectors. Unlike conventional methods relying on predefined molecular descriptors, GNNs learn directly from molecular structures, reducing feature selection biases. By enhancing the interpretability of AI-driven predictions, GNNs improve understanding of decision-making. To construct GNN-based predictive models, we compiled datasets of compounds classified as either bitter or sweet. Our models achieved prediction accuracies comparable to or exceeding those of traditional machine learning and deep learning models that rely on molecular descriptors. To enhance model interpretability, we employed the Integrated Gradients method to visualize the molecular features influencing bitterness or sweetness predictions. These visualizations were further validated through molecular docking simulations of ligands on taste receptors, using the AlphaFold Protein Structure Database. Bitterness was evaluated using TAS2R16 and sweetness with TAS1R2. The chemical visualization results were then compared with conformational data, demonstrating strong alignment with previous experimental and computational analyses. These findings validate our AI model's accuracy and visualization outcomes, highlighting the potential of GNN-based models in taste prediction. This approach offers a novel framework for understanding the molecular mechanisms underlying taste perception. Further investigations are warranted to explore these mechanisms in greater depth and extend this methodology to predict additional taste modalities. |
| format | Article |
| id | doaj-art-c9ab47c59c334f618b6ded2013662890 |
| institution | DOAJ |
| issn | 2665-9271 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Current Research in Food Science |
| spelling | doaj-art-c9ab47c59c334f618b6ded20136628902025-08-20T03:20:58ZengElsevierCurrent Research in Food Science2665-92712025-01-011010109010.1016/j.crfs.2025.101090AI-driven prediction of bitterness and sweetness and analysis of receptor interactionsHiroaki Iwata0Department of Biological Regulation, Faculty of Medicine, Tottori University, 86 Nishi-cho, Yonago 683-8503, JapanUnderstanding the molecular mechanisms governing sweetness and bitterness is essential for identifying desirable taste characteristics in natural and synthetic compounds. In this study, we developed graph neural network (GNN)-based artificial intelligence (AI) models to predict bitterness and sweetness based on chemical structure. GNNs utilize deep learning to capture relationships among molecular components within a graph, extracting latent molecular vectors. Unlike conventional methods relying on predefined molecular descriptors, GNNs learn directly from molecular structures, reducing feature selection biases. By enhancing the interpretability of AI-driven predictions, GNNs improve understanding of decision-making. To construct GNN-based predictive models, we compiled datasets of compounds classified as either bitter or sweet. Our models achieved prediction accuracies comparable to or exceeding those of traditional machine learning and deep learning models that rely on molecular descriptors. To enhance model interpretability, we employed the Integrated Gradients method to visualize the molecular features influencing bitterness or sweetness predictions. These visualizations were further validated through molecular docking simulations of ligands on taste receptors, using the AlphaFold Protein Structure Database. Bitterness was evaluated using TAS2R16 and sweetness with TAS1R2. The chemical visualization results were then compared with conformational data, demonstrating strong alignment with previous experimental and computational analyses. These findings validate our AI model's accuracy and visualization outcomes, highlighting the potential of GNN-based models in taste prediction. This approach offers a novel framework for understanding the molecular mechanisms underlying taste perception. Further investigations are warranted to explore these mechanisms in greater depth and extend this methodology to predict additional taste modalities.http://www.sciencedirect.com/science/article/pii/S2665927125001212Taste predictionGraph neural networkBitterness/sweetness taste receptorDocking simulationExplainable AITaste perception |
| spellingShingle | Hiroaki Iwata AI-driven prediction of bitterness and sweetness and analysis of receptor interactions Current Research in Food Science Taste prediction Graph neural network Bitterness/sweetness taste receptor Docking simulation Explainable AI Taste perception |
| title | AI-driven prediction of bitterness and sweetness and analysis of receptor interactions |
| title_full | AI-driven prediction of bitterness and sweetness and analysis of receptor interactions |
| title_fullStr | AI-driven prediction of bitterness and sweetness and analysis of receptor interactions |
| title_full_unstemmed | AI-driven prediction of bitterness and sweetness and analysis of receptor interactions |
| title_short | AI-driven prediction of bitterness and sweetness and analysis of receptor interactions |
| title_sort | ai driven prediction of bitterness and sweetness and analysis of receptor interactions |
| topic | Taste prediction Graph neural network Bitterness/sweetness taste receptor Docking simulation Explainable AI Taste perception |
| url | http://www.sciencedirect.com/science/article/pii/S2665927125001212 |
| work_keys_str_mv | AT hiroakiiwata aidrivenpredictionofbitternessandsweetnessandanalysisofreceptorinteractions |