Protein powder, Parmesan, and kefir sometimes leave a bitter aftertaste—more often than not it’s the bitter peptides that show up when proteins are broken down and cause trouble.
At the Leibniz Institute for Food Systems Biology at TUM, the team combines a protein language model with their own BitterPep-GCN. They let AI predict whether peptides are bitter, and even design entirely new peptides from scratch. First, they generate 161 new sequences, then select the 31 most promising candidates for synthesis. Next, a trained sensory panel tastes them: 25 are correctly classified (15 bitter, 10 not bitter).
It’s another step closer to “getting a handle on the weird off-flavors of plant proteins and fermented foods in advance.” The paper is published in npj Science of Food.
#AI #Food Science
At the Leibniz Institute for Food Systems Biology at TUM, the team combines a protein language model with their own BitterPep-GCN. They let AI predict whether peptides are bitter, and even design entirely new peptides from scratch. First, they generate 161 new sequences, then select the 31 most promising candidates for synthesis. Next, a trained sensory panel tastes them: 25 are correctly classified (15 bitter, 10 not bitter).
It’s another step closer to “getting a handle on the weird off-flavors of plant proteins and fermented foods in advance.” The paper is published in npj Science of Food.
#AI #Food Science
