Breakthrough Tracker record
Natural-language protein designs show four intended functions in laboratory tests
Dai, You, Yuan and collaborators report that Pinal, a 16-billion-parameter text-to-protein model, generated candidates with experimentally observed function in four tasks: fluorescence, PET hydrolysis, alcohol dehydrogenation and metabolic H-protein activity.
← Back to the filtered Breakthrough Tracker
- Stable ID
biology-pinal-natural-language-protein-design-2026- Revision
biology-pinal-natural-language-protein-design-2026.v1- Field
- Biology and artificial intelligence · Protein design, molecular engineering and generative biology
- Evidence
- Tier 1 · Peer reviewed: Yes
- Record state
- Current · Peer-reviewed laboratory validation
- Last checked
AI role
Pinal is the central generative system. It was trained on 1.7 billion synthetically annotated protein–text pairs and maps functional descriptions through structure to sequence; wet-lab assays, rather than model scores alone, supply the decisive evidence.
Record details
- Problem or result
- Functional protein candidate generation from natural-language specifications
- Authors
- Fengyuan Dai, Shiyang You, Yudian Zhu, Fajie Yuan and collaborators
- Institutions
- Westlake University-led multi-institution collaboration; complete affiliations are listed in the Nature record
- Result date
- Nature publication October 21, 2026; underlying preprint revised before publication
Why it matters
The work moves text-conditioned protein design beyond predicted structure by testing distinct functions in the laboratory. A designed H-protein increased product titre 1.7-fold over the native Escherichia coli H-protein in a multi-enzyme carbon-fixation pathway without iterative experimental optimization.
Limits
Four successful task classes do not establish general prompt-to-protein reliability, safety or superiority to specialist design pipelines. Most candidate screening remained computational, and the reported H-protein gain is a pathway-specific laboratory result rather than an organism-level or industrial production result.
Sources
- Primary: Nature Pinal protein-design study
- Code and model artifact: Pinal public implementation and model links
- Preprint: Pinal bioRxiv preprint, version 2
Correction and revision history
- 2026-10-22 — Added after primary-source, scope, status, AI-role and limitation review.
Machine-readable: JSON v1 · CSV v1 · Schema v1
This individual record remains noindex until a story-specific featured image passes Kingy’s rendered-pixel visual review. The source-linked tracker hub remains the public index.