Enzymes designed with the help of generative artificial-intelligence models can break down a widely used packaging plastic in a matter of hours, according to results presented at a conference this week. The designed proteins outperform the best natural enzymes by more than tenfold at moderate temperatures.
The team used a model trained on millions of protein structures to propose candidate sequences, then tested several hundred in the laboratory. "The hit rate was much higher than we expected," says Kenji Sato, a protein engineer at Tsukishima University.
Independent researchers cautioned that scaling up enzymatic recycling poses engineering challenges, but called the results a striking demonstration of design.
