UmamiPredict: machine learning model to predict umami taste of molecules and peptides

Area of Science:

• Food Science and Technology
• Computational Chemistry
• Machine Learning

Background:

• Umami, the fifth basic taste, is triggered by amino acids and nucleotides like L-glutamate.
• Traditional umami-rich foods include soy sauce, cheese, and fermented products.
• Predicting umami taste computationally is challenging due to limited data and molecular feature representation.

Purpose of the Study:

• To develop a computational model for classifying peptides and small molecules as umami or non-umami.
• To address the lack of datasets and inadequate feature representation in existing methods.
• To provide a user-friendly tool for predicting umami taste.

Main Methods:

• Curated a balanced dataset of 868 compounds (439 umami, 429 non-umami).
• Extracted molecular descriptors for physicochemical and structural properties.
• Employed ensemble machine learning models like LightGBM, XGBoost, and ExtraTrees.

Main Results:

• Random forest achieved 92.13% accuracy on peptides; LDA and ExtraTrees reached 98.84% on small molecules.
• LightGBM model attained 96.55% accuracy on the combined dataset.
• Developed UmamiPredict web server for user-friendly molecule umami taste prediction.

Conclusions:

• Machine learning effectively predicts umami taste from molecular structures.
• Integrating peptide and small molecule data enhances prediction accuracy.
• UmamiPredict serves as a valuable tool for researchers in food science and beyond.