Programmable design of functional proteins from natural language
Published in Nature, 2026
16B-parameter model for designing functional proteins from natural language, with experimental validation across four protein tasks.
Published in Nature on 21 October 2026.
Pinal uses natural-language instructions to generate structures and then design protein sequences. Its 16B parameters were trained on 1.7 billion synthetically annotated protein-text pairs. Experimental tests confirmed the requested functions for a fluorescent protein, a PET hydrolase, an alcohol dehydrogenase, and a metabolic H-protein.
In a multi-enzyme CO2 fixation pathway, a designed H-protein delivered 1.7 times the product titre obtained with the corresponding native E. coli H-protein, without iterative experimental optimization.
* Fengyuan Dai and Shiyang You contributed equally.