Integrated AI Framework Proposed for Predicting, Debittering, and Evaluating Bitter Peptides

While bioactive peptides (BAPs) offer significant health benefits, their widespread industrial application in functional foods and nutraceuticals is severely hampered by their intense bitterness. Current methods for identifying and mitigating this bitterness are often inefficient or lack precision. Moreover, recent research suggests that some bitter compounds may positively influence gastrointestinal tract hormones and satiety, adding complexity to their perception and utility. Addressing this bitterness barrier is crucial for unlocking the full potential of BAPs.

This review systematically constructed a collaborative theoretical framework for bitter peptides (BPs), integrating intelligent prediction, targeted debittering, and multi-dimensional evaluation. The authors first examined the application of deep learning techniques, such as quantitative structure-activity relationship (QSAR) and graph convolutional network (GCN), alongside molecular docking for high-throughput BP identification. Subsequently, they explored how artificial intelligence and computational simulation can enhance traditional debittering processes, focusing on multifunctional composite wall materials for encapsulation and microbial fermentation for metabolic regulation. Finally, a three-dimensional cross-validation system, combining standardized quantitative sensory evaluation and biomimetic electronic tongues, was established to provide high-fidelity data for AI models.

The review highlights a comprehensive strategy to address the challenge of bitter peptides (BPs). It details how deep learning, specifically QSAR and GCN, coupled with molecular docking, can effectively identify BPs at high throughput and elucidate their target receptor interaction mechanisms. The authors found that AI and computational simulation significantly improve the efficiency of debittering processes, particularly through the use of multifunctional composite wall materials for targeted encapsulation and delivery of BPs. Furthermore, they emphasized the metabolic regulatory mechanisms behind controlling microbial fermentation for the debittering of specific peptide substrates. To ensure robust AI model development, a novel three-dimensional cross-validation system was established, integrating standardized quantitative sensory evaluation with biomimetic electronic tongues. This system provides a crucial data closed loop for refining predictive models.

Future research should focus on developing large models for flavor generation to drive the green and targeted creation of low-bitterness and highly active peptides.