Natural Language Protein Design Tool Pinal Released

According to an Ars Technica report on October 28, 2026, a Chinese research team has released a tool called Pinal that converts natural-language text descriptions into protein sequences. Its workflow has two steps: first, the T2struct module converts the prompt into a protein structure blueprint, then the SAProt-T module fills in the amino acid sequence based on the blueprint. The model's training data includes 800 million keyword-protein pairs and uses structures predicted by AlphaFold. In the example given in the report, the team used Pinal to generate 50,000 candidate sequences of GFP-like fluorescent proteins, computationally screened out 29 for synthesis, and ultimately only 1 showed actual fluorescence, revealing a significant gap between computational design and experimental validation. Ars Technica discussed Pinal alongside another effort bringing natural language into genome design (the Carbon X model on Hugging Face), interpreting such work as a trend of natural-language interfaces entering the field of biological design. There have been no further reports on whether the tool is open source, the team's name, or additional experimental validation results.

AI overview of reporting · Updated 2 h前

Not enough continuous observations to show a trend yet.