Briefs · Emerging Technologies
Scientists Can Now Ask AI to Design Proteins Using Plain Language
Protein design just got a plain-English interface. No PhD in molecular origami required.
Pinal, an AI system from Westlake University in China published in Nature, turns a sentence describing what a protein should do into an amino acid sequence — no structural constraints, no specialist software. It works in two stages: one module sketches a structural blueprint from the words, another picks the amino acids. Trained on roughly 135 million protein sequences, it designed four test proteins, including a PET-plastic-eating enzyme and a shuttle protein that boosted a carbon-dioxide fixation pathway to 1.7 times the natural output — though the glowing protein took 50,000 tries to get one that actually glowed.
Meanwhile Carbon-X, an open model from Hugging Face and collaborators, chats with DNA and runs on a MacBook.
Conclusion: biology is not a members-only club any more. It is a group chat.
Read the original at Singularity Hub
Summary in our own words. The original belongs to its publisher.