University of Washington researchers published a model that accepts protein design goals written in natural language. Unlike many AI-for-science announcements, there is lab data: nine designed proteins bound to a rare-disease target. There are also clear limits.
Prompteina is a generative model from the UW Medicine Institute for Protein Design team (including Nobel laureate David Baker) in collaboration with natural language researchers, among them Hanna Hajishirzi (Allen Institute for AI), Noah A. Smith, and Allyson Ettinger, according to Tech Times. The work appeared on arXiv on October 9, 2026 and was accepted to the NeurIPS 2026 main conference. The code and data are published on GitHub.
The architecture is flow matching, inherited from Proteina-Complexa, and it generates complete three-dimensional structures with all-atom coordinates. The novelty is that the goal is expressed in English: for example, the kind of binding sought against a target. To achieve this, the team trained on 1.8 million protein structures paired with more than 50 million automatically generated descriptions, and added text-residue alignment during fine-tuning. In essence, it replaces rigid technical specifications with a textual description of the objective.
This is far stronger than a simulation: the binders were measured in laboratory experiments. But an affinity of 107 nanomolar sits above the typical therapeutic range of 1 to 10 nM, so these are starting points, not drugs.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel