Taste is crucial in driving food choice and preference. Umami is one of the basic tastes defined by characteristic deliciousness and mouthfulness that it imparts to foods. Identification of ingredients to enhance umami taste is of significant value to food industry. Various models have been shown to predict umami taste using feature encodings derived from traditional molecular descriptors such as amphiphilic pseudo-amino acid composition, dipeptide composition, and composition-transition-distribution. Highest reported accuracy of 90.5 % was recently achieved through novel model architecture. Here, we propose use of biological sequence transformers such as ProtBert and ESM2, trained on the Uniref databases, as the feature encoders block. With combination of 2 encoders and 2 classifiers, 4 model architectures were developed. Among the 4 models, ProtBert-CNN model outperformed other models with accuracy of 95 % on 5-fold cross validation data and 94 % on independent data.

Sweetness in food delivers a delightful sensory experience, underscoring the crucial role of sweeteners in the food industry. However, the widespread use of sweeteners has sparked health concerns. This underscores the importance of developing and screening natural, health-conscious sweeteners. Our study represents a groundbreaking venture into the discovery of such sweeteners derived from egg and soy proteins. Employing virtual hydrolysis as a novel technique, our research entailed a comprehensive screening process that evaluated biological activity, solubility, and toxicity of the derived compounds. We harnessed cutting-edge machine learning methodologies, specifically the latest graph neural network models, for predicting the sweetness of molecules. Subsequent refinements were made through molecular docking screenings and molecular dynamics simulations. This meticulous research approach culminated in the identification of three promising sweet peptides: DCY(Asp-Cys-Tyr), GGR(Gly-Gly-Arg), and IGR(Ile-Gly-Arg). Their binding affinity with T1R2/T1R3 was lower than -15 kcal/mol. Using an electronic tongue, we verified the taste profiles of these peptides, with IGR emerging as the most favorable in terms of taste with a sweetness value of 19.29 and bitterness value of 1.71. This study not only reveals the potential of these natural peptides as healthier alternatives to traditional sweeteners in food applications but also demonstrates the successful synergy of computational predictions and experimental validations in the realm of flavor science.

Bitter taste involves the detection of diverse chemical compounds by a family of G protein-coupled receptors, known as taste receptor type 2 (TAS2R). It is often linked to toxins and harmful compounds and in particular bitter taste receptors participate in the regulation of glucose homeostasis, modulation of immune and inflammatory responses, and may have implications for various diseases. Human TAS2Rs are characterized by their polymorphism and differ in localization and function. Different receptors can activate various signaling pathways depending on the tissue and the ligand. However, in vitro screening of possible TAS2R ligands is costly and time-consuming. For this reason, in silico methods to predict bitterant-TAS2R interactions could be powerful tools to help in the selection of ligands and targets for experimental studies and improve our knowledge of bitter receptor roles. Machine learning (ML) is a branch of artificial intelligence that applies algorithms to large datasets to learn from patterns and make predictions. In recent years, there has been a record of numerous taste classifiers in literature, especially on bitter/non-bitter or bitter/sweet classification. However, only a few of them exploit ML to predict which TAS2R receptors could be targeted by bitter molecules. Indeed, the shortage and incompleteness of data on receptor-ligand associations in literature make this task non-trivial. In this work, we provide an overview of the state of the art dealing with this specific investigation, focusing on three ML-based models, namely BitterX (2016), BitterSweet (2019) and BitterMatch (2022). This review aims to establish the foundation for future research endeavours focused on addressing the limitations and drawbacks of existing models.

Taste determination in small molecules is critical in food chemistry but traditional experimental methods can be time-consuming. Consequently, computational techniques have emerged as valuable tools for this task. In this study, we explore taste prediction using various molecular feature representations and assess the performance of different machine learning algorithms on a dataset comprising 2601 molecules. The results reveal that GNN-based models outperform other approaches in taste prediction. Moreover, consensus models that combine diverse molecular representations demonstrate improved performance. Among these, the molecular fingerprints + GNN consensus model emerges as the top performer, highlighting the complementary strengths of GNNs and molecular fingerprints. These findings have significant implications for food chemistry research and related fields. By leveraging these computational approaches, taste prediction can be expedited, leading to advancements in understanding the relationship between molecular structure and taste perception in various food components and related compounds.

Sweetness is a vital taste to which humans are innately attracted. Given the increasing prevalence of type-2 diabetes, it is highly relevant to build computational models to predict the sweetness of small molecules. Such models are valuable for identifying sweeteners with low calorific value. We present regression-based machine learning and deep learning algorithms for predicting sweetness. Toward this goal, we manually curated the most extensive dataset of 671 sweet molecules with known experimental sweetness values ranging from 0.2 to 22,500,000. Gradient Boost and Random Forest Regressors emerged as the best models for predicting the sweetness of molecules with a correlation coefficient of 0.94 and 0.92, respectively. Our models show state-of-the-art performance when compared with previously published studies. Besides making our dataset (SweetpredDB) available, we also present a user-friendly web server to return the predicted sweetness for small molecules, Sweetpred (https://cosylab.iiitd.edu.in/sweetpred).

Taste is a sensory modality crucial for nutrition and survival, since it allows the discrimination between healthy foods and toxic substances thanks to five tastes, i.e., sweet, bitter, umami, salty, and sour, associated with distinct nutritional or physiological needs. Today, taste prediction plays a key role in several fields, e.g., medical, industrial, or pharmaceutical, but the complexity of the taste perception process, its multidisciplinary nature, and the high number of potentially relevant players and features at the basis of the taste sensation make taste prediction a very complex task. In this context, the emerging capabilities of machine learning have provided fruitful insights in this field of research, allowing to consider and integrate a very large number of variables and identifying hidden correlations underlying the perception of a particular taste. This review aims at summarizing the latest advances in taste prediction, analyzing available food-related databases and taste prediction tools developed in recent years.
UNASSIGNED: The online version contains supplementary material available at 10.1007/s00217-022-04044-5.

Taste of a chemical compound present in food stimulates us to take in nutrients and avoid poisons. However, the perception of taste greatly depends on the genetic as well as evolutionary perspectives. The aim of this work was the development and validation of a machine learning model based on molecular fingerprints to discriminate between sweet and bitter taste of molecules. BitterSweetForest is the first open access model based on KNIME workflow that provides platform for prediction of bitter and sweet taste of chemical compounds using molecular fingerprints and Random Forest based classifier. The constructed model yielded an accuracy of 95% and an AUC of 0.98 in cross-validation. In independent test set, BitterSweetForest achieved an accuracy of 96% and an AUC of 0.98 for bitter and sweet taste prediction. The constructed model was further applied to predict the bitter and sweet taste of natural compounds, approved drugs as well as on an acute toxicity compound data set. BitterSweetForest suggests 70% of the natural product space, as bitter and 10% of the natural product space as sweet with confidence score of 0.60 and above. 77% of the approved drug set was predicted as bitter and 2% as sweet with a confidence score of 0.75 and above. Similarly, 75% of the total compounds from acute oral toxicity class were predicted only as bitter with a minimum confidence score of 0.75, revealing toxic compounds are mostly bitter. Furthermore, we applied a Bayesian based feature analysis method to discriminate the most occurring chemical features between sweet and bitter compounds using the feature space of a circular fingerprint.

In-silico bitterant prediction received the considerable attention due to the expensive and laborious experimental-screening of the bitterant. In this work, we collect the fully experimental dataset containing 707 bitterants and 592 non-bitterants, which is distinct from the fully or partially hypothetical non-bitterant dataset used in the previous works. Based on this experimental dataset, we harness the consensus votes from the multiple machine-learning methods (e.g., deep learning etc.) combined with the molecular fingerprint to build the bitter/bitterless classification models with five-fold cross-validation, which are further inspected by the Y-randomization test and applicability domain analysis. One of the best consensus models affords the accuracy, precision, specificity, sensitivity, F1-score, and Matthews correlation coefficient (MCC) of 0.929, 0.918, 0.898, 0.954, 0.936, and 0.856 respectively on our test set. For the automatic prediction of bitterant, a graphic program \"e-Bitter\" is developed for the convenience of users via the simple mouse click. To our best knowledge, it is for the first time to adopt the consensus model for the bitterant prediction and develop the first free stand-alone software for the experimental food scientist.

The role of bitter taste-one of the few basic taste modalities-is commonly assumed to signal toxicity and alert animals against consuming harmful compounds. However, it is known that some toxic compounds are not bitter and that many bitter compounds have negligible toxicity while having important health benefits. Here we apply a quantitative analysis of the chemical space to shed light on the bitterness-toxicity relationship. Using the BitterDB dataset of bitter molecules, The BitterPredict prediction tool, and datasets of toxic compounds, we quantify the identity and similarity between bitter and toxic compounds. About 60% of the bitter compounds have documented toxicity and only 56% of the toxic compounds are known or predicted to be bitter. The LD50 value distributions suggest that most of the bitter compounds are not very toxic, but there is a somewhat higher chance of toxicity for known bitter compounds compared to known nonbitter ones. Flavonoids and alpha acids are more common in the bitter dataset compared with the toxic dataset. In contrast, alkaloids are more common in the toxic datasets compared to the bitter dataset. Interestingly, no trend linking LD50 values with the number of activated bitter taste receptors (TAS2Rs) subtypes is apparent in the currently available data. This is in accord with the newly discovered expression of TAS2Rs in several extra-oral tissues, in which they might be activated by yet unknown endogenous ligands and play non-gustatory physiological roles. These results suggest that bitter taste is not a very reliable marker for toxicity, and is likely to have other physiological roles. © 2017 IUBMB Life, 69(12):938-946, 2017.