Pinal on Nature: Designing Functional Proteins Directly from Natural Language
bravo_abad · x · 2026-10-22
Dai et al. published Pinal, a generative protein-design model, in Nature: scientists describe a desired protein function in plain language (e.g., a plastic-degrading enzyme) without supplying structural constraints, and the model proposes structure and sequence.
How it works:
- A Transformer-based architecture first generates a compact description of a candidate protein structure from the text
- A second model uses that description plus the original request to generate the amino acid sequence
- Structure serves as a useful bridge: proteins with very different sequences can adopt similar shapes
The team also validated beyond single-enzyme design, including an H-protein that shuttles reaction intermediates between enzymes, demonstrating a full pipeline from natural-language function to experimentally tested proteins.
More from Research
- Deep dive: running RL on heterogeneous hardware — Blackwells for training, TPUs for inference — A_K_Nain · 2026-10-22
- Long Now Labs picks first awardees, including a thermodynamic clock experiment for LLMs — begusgasper · 2026-10-22
- Princeton opens AI Postdoc fellowship in Natural and Artificial Minds, deadline Feb 28 — LakeBrenden · 2026-10-22
- ESMFold2 API adds max accuracy mode after single-sequence precision complaints — proteinrosh · 2026-10-22
- OpenTac team to demo tactile sensors at CoRL with small-scale sensor giveaway — anand_bhattad · 2026-10-22
- Marathon Fusion claims scalable mercury-to-gold transmutation in fusion plants — clarejtbirch · 2026-10-22