Toward a general and interpretable umami taste predictor using a multi-objective machine learning approach.

Area of Science:

• Food Science
• Computational Chemistry
• Sensory Science

Background:

• Umami taste is a fundamental taste modality linked to protein content.
• Predicting umami taste is crucial for food science and ingredient development.
• Existing prediction tools are limited to peptide amino acid sequences.

Purpose of the Study:

• To develop a novel, broadly applicable machine learning (ML) algorithm for umami taste prediction.
• To enable umami taste prediction from a molecule's standard chemical representation.
• To facilitate the rational design of compounds with specific taste profiles.

Main Methods:

• Developed a machine learning-based algorithm named VirtuousUmami.
• Utilized the Simplified Molecular Input Line Entry System (SMILES) representation for molecular input.
• Tested the model on five diverse databases of food-related compounds and natural products.

Main Results:

• VirtuousUmami successfully predicts umami taste from SMILES representations.
• The model demonstrates applicability across various food and natural compound databases.
• This approach broadens the scope of umami taste prediction beyond peptides.

Conclusions:

• The VirtuousUmami algorithm offers a versatile tool for umami taste prediction.
• This work advances the understanding of molecular features governing umami taste.
• The tool supports the rational design of novel umami-enhancing compounds and ingredients.