A research team from the Leibniz Institute for Food Systems Biology at the Technical University of Munich has developed a new AI-based method that can predict the bitterness of peptides—small protein fragments. This method also enables the design of new bitter-tasting peptides from scratch. The findings were published in the journal npj Science of Food. Bitter-tasting peptides are created when proteins are broken down by enzymes or chemical processes. These peptides are a challenge in the production of foods like fermented products or protein hydrolysates, which are used in various food applications. However, they can also have health-related effects, such as influencing feelings of hunger and fullness. Understanding and controlling their taste is important for improving the sensory experience of food. The researchers combined a protein language model, trained using around 500 known bitter peptides, with a specialized type of artificial intelligence called BitterPep-GCN. This model uses a graph convolutional network (GCN), a form of artificial neural network designed to analyze complex, structured data. Using this combined system, the team generated 161 new peptide sequences that had not yet been tested in experiments. Based on the model’s predictions, they selected the most promising candidates and tested them using a trained sensory panel. Out of 31 tested peptides, the panel correctly identified 25 as either bitter or non-bitter. The study also revealed several previously unknown bitter and non-bitter peptides. Alexandra Steuer, the study’s first author and a doctoral student, noted that the AI method not only predicts bitterness but also allows for the specific design of new bitter peptides. Antonella Di Pizio, the principal investigator, said the research is now ready for practical application. In the long term, this could help control the formation of bitter peptides in food production, particularly in plant-based, protein-rich foods, which often face challenges due to undesirable flavors.