AI Designs New Proteins
- Reports this week highlighted AI systems being used to design novel protein sequences. - The work produced protein designs not seen in nature, according to coverage of recent studies. - Media framed this as part of a cluster of biotech advances, alongside bacterial DNA editing and femtosecond laser control experiments. (x.com)
Proteins are chains of amino acids that act as tiny machines in cells, and this week labs used AI to design protein sequences not found in nature. (nature.com) A team led by Gyu Rie Lee and collaborators published a Nature Communications paper on 28 March 2026 that used deep learning plus physics-based design to create binders for six small-molecule targets. (nature.com) That Nature Communications study reported nanomolar to low-micromolar binding affinities and used a cortisol binder to build a chemically induced dimerization biosensor. (nature.com) On 24 March 2026 MIT researchers published a Matter paper describing VibeGen, a generative AI that designs proteins by prescribing how they should flex and vibrate rather than only fixing static shapes. (news.mit.edu) On 15 April 2026 an international team led by Kirill Alexandrov and coauthors published in Nature Biotechnology artificial allosteric protein switches that worked inside Escherichia coli and in bioelectronic sensors. (nature.com) Taken together, these papers illustrate AI-driven de novo design that produces sequences and binding pockets not present in natural databases; the Nature Communications team deposited crystal structures (PDB IDs 8UQF, 8VFQ, 8VEZ) and released design scripts on Zenodo. (nature.com) The shift builds on earlier advances from structure prediction to generative design — David Baker and DeepMind researchers were honored by the 2024 Nobel Prize in Chemistry for work on computational protein design and structure prediction. (nobelprize.org) Authors and funders emphasize applications in diagnostics, drug discovery and environmental sensing, while biosecurity researchers note risks: a Science study led by Microsoft in October 2025 found AI-generated protein variants could evade some DNA-synthesis screening tools until providers deployed targeted patches. (qut.edu.au) (pubmed.ncbi.nlm.nih.gov) The immediate next steps are experimental scaling and scrutiny — teams have released data and code for independent validation, and DNA-synthesis providers and researchers are updating screening and oversight as these AI-designed sequences move from computer to lab. (nature.com)
