BIPE: Artificial Intelligence-Driven Peptide Bitterness Intensity Prediction Engine.

Area of Science:

• Food Science
• Biotechnology
• Computational Biology

Background:

• Bitterness is a key taste, but bitter peptides from protein processing cause consumer rejection despite potential bioactivities.
• Accurate bitterness assessment is crucial for product development but current methods are costly and slow.
• A need exists for efficient tools to predict and manage peptide bitterness.

Purpose of the Study:

• To develop an accurate and efficient computational model for predicting peptide bitterness intensity from amino acid sequence.
• To enable early-stage assessment of bitterness in food protein development and process optimization.
• To provide mechanistic insights into the amino acid composition underlying peptide bitterness.

Main Methods:

• Developed BIPE (Bitterness Intensity Prediction Engine), an end-to-end regression model.
• Integrated ESM3 protein language model representations with a multilayer perceptron readout.
• Performed regression of bitterness thresholds in log space using sequence data.

Main Results:

• BIPE achieved high accuracy (R²=0.9050 cross-validation, R²=0.9449 independent test set).
• The model demonstrated external validity by correlating with electronic tongue and human sensory data.
• BIPE successfully differentiated bitterness in soybean protein hydrolysates and revealed amino acid patterns linked to bitterness.

Conclusions:

• BIPE provides a quantitative and sequence-based method for assessing peptide bitterness intensity.
• The model facilitates the rational design of low-bitterness peptides for improved food products.
• BIPE serves as a baseline for taste modeling and supports flavor engineering and process optimization.