Designing Protein Structures from Natural Language
Prompteina turns plain-language design goals into all-atom protein structures, including experimentally validated binders.
The paper introduces Prompteina, trained on 1.8M structures and more than 50M captions tying text to structural and functional protein properties. It beats baselines on most tested structural properties and produces proteins that better match intended functional descriptions. In binder design, it leads on at least 13 of 20 targets and improves interface confidence over the strongest baseline. The authors report nine validated binders against GM2A, with the strongest showing an apparent KD of 107 nM. ArXiv · AI/CL/LG's note
The paper introduces Prompteina, trained on 1.8M structures and more than 50M captions tying text to structural and functional protein properties. It beats baselines on most tested structural properties and produces proteins that better match intended functional descriptions. In binder design, it leads on at least 13 of 20 targets and improves interface confidence over the strongest baseline. The authors report nine validated binders against GM2A, with the strongest showing an apparent KD of 107 nM. ArXiv · AI/CL/LG's note
score 6