Machine learning-driven classification of bioactive peptides and their molecular basis for sour taste perception

Area of Science:

• Food Science
• Biotechnology
• Computational Chemistry

Background:

• Dairy proteins yield bioactive peptides with sensory and health benefits.
• Yogurt proteolysis produces peptides, including sour-taste peptides, that can improve flavor.
• Current methods for identifying these peptides are time-consuming and costly.

Purpose of the Study:

• To develop an integrated computational-experimental workflow for identifying yogurt-derived sour peptides.
• To discover peptides with potential multifunctional applications in flavor and health.
• To establish a rapid and cost-effective peptide discovery strategy.

Main Methods:

• Curated a dataset of sour and non-sour peptides for machine learning model training.
• Employed machine learning models (XGBoost, Random Forest, Multi-Layer Perceptron) to predict sourness.
• Synthesized top peptide candidates and evaluated them using an electronic tongue and molecular dynamics simulations.
• Assessed peptide interactions with the OTOP1 proton channel.

Main Results:

• XGBoost model achieved high predictive performance (AUC >0.95) in identifying sour peptides.
• DSEPV was confirmed as a potent sour-taste peptide; others showed balanced taste profiles.
• Molecular simulations indicated stable peptide-protein complexes with OTOP1, suggesting potential bioactivity.
• The workflow proved rapid and cost-effective for peptide discovery.

Conclusions:

• An integrated computational-experimental approach enables efficient discovery of flavor-modifying peptides from dairy.
• Yogurt-derived peptides like DSEPV offer potential for natural flavor enhancement and functional food development.
• This strategy supports sustainable upcycling of fermentation by-products for dual-purpose applications.