Abstract:Umami, a fundamental human taste modality, refers to the savory flavors in meats and broths, often associated with monosodium glutamate and protein richness. With limited knowledge of umami molecules, the food industry seeks efficient approaches for identifying novel tastants. In this study, we have devised a virtual screening pipeline for identifying potential novel umami tastants from molecular databases. We first curated a comprehensive classification dataset containing 439 umami and 428 non-umami molecules. A transformer-based architecture was trained to differentiate between the two classes, achieving the best performance to date. Additionally, we built a neural network model for predicting the potency of umami compounds, the first effort of its kind. These two models, in conjunction with similarity analysis and toxicity screening, form an end-to-end framework for the rational discovery of novel tastants. We finally applied this framework to the FooDB database as an illustrative use case. This study demonstrates the potential of data-driven methods in predicting the taste of molecules from structural and chemical features.

What problem does this paper attempt to address?

This paper mainly discusses the use of deep learning to screen umami taste substances. Umami is one of the basic taste patterns of human beings, which is usually associated with the savory taste in meat and soup and is related to high levels of monosodium glutamate and protein. Due to limited understanding of umami molecules, the food industry needs more effective methods to find new umami substances. In the study, the authors constructed a virtual screening pipeline to identify potential new umami substances from molecular databases. They first created a classification dataset containing 439 umami molecules and 428 non-umami molecules, and trained a model based on Transformer architecture to distinguish between the two classes of molecules, achieving the best performance to date. In addition, they developed a neural network model to predict the intensity of umami compounds, which is the first attempt of its kind. These two models, combined with similarity analysis and toxicity screening, form a complete end-to-end framework for the rational discovery of new umami substances. The paper demonstrates the effectiveness of this framework using the FooDB database as an example application. The study highlights the potential of data-driven approaches in predicting taste based on molecular structure and chemical characteristics, and notes that despite existing models for sweet and bitter tastes, relatively less work has been done in predicting umami taste. In the paper, the authors constructed the largest umami classification dataset to date and used deep learning techniques for molecule classification and intensity prediction, aiming to improve the efficiency and accuracy of compound taste prediction in the food industry.

Computational Screening of Umami Tastants Using Deep Learning

Computational screening of umami tastants using deep learning

A Novel Bionic in Vitro Bioelectronic Tongue Based on Cardiomyocytes and Microelectrode Array for Bitter and Umami Detection.

Design of virtual libraries of umami-tasting molecules

VmmScore: An umami peptide prediction and receptor matching program based on a deep learning approach

UmamiPreDL: Deep learning model for umami taste prediction of peptides using BERT and CNN

Identification of novel umami molecules via QSAR models and molecular docking

Electronic Tongues for Inedible Media

Bimodal Fusion Network for Basic Taste Sensation Recognition from Electroencephalography and Electromyography

A Machine Learning Method to Identify Umami Peptide Sequences by Using Multiplicative LSTM Embedded Features

A TastePeptides-Meta System Including an Umami/bitter Classification Model Umami_YYDS, a TastePeptidesDB Database and an Open-Source Package Auto_Taste_ML

TastePeptides-EEG: an Ensemble Model for Umami Taste Evaluation Based on Electroencephalogram and Machine Learning.

A Chemical Language Model for Molecular Taste Prediction

In-silico Investigation of Umami Peptides with Receptor T1R1/T1R3 for the Discovering Potential Targets: A Combined Modeling Approach

Rapid Screening of Novel Umami Peptides in High-Umami Scored Air-Dried Chicken under Low-Salt Processing Based on Molecular Sensory Techniques

Integration of cloud-based molecular networking and docking for enhanced umami peptide screening from Pixian douban

A Chemical Language Model for Multi-Class Molecular Taste Prediction

Peptidomics Screening and Molecular Docking with Umami Receptors T1R1/T1R3 of Novel Umami Peptides from Oyster (Crassostrea gigas) Hydrolysates

Umami-MRNN: Deep learning-based prediction of umami peptide using RNN and MLP

Umami Taste Evaluation Based on a Novel Mouse Taste Receptor Cell-Based Biosensor

A Rational Tool for the Umami Evaluation of Peptides Based on Multi-Techniques.