Pinal designs proteins from plain-language prompts

Researchers in China unveiled Pinal, a system that turns natural-language descriptions into protein sequences with a specified function. A T2struct module converts the prompt into a structural blueprint, and SAProt-T builds the amino acid sequence. Training used databases with hundreds of millions of keyword–protein pairs plus AlphaFold structures.

• Pinal takes plain text instead of precise amino acid coordinates
• T2struct builds a structural blueprint; SAProt-T assembles the sequence
• Training used databases, one with 800 million keyword–protein pairs
• The system was tested on four different types of proteins