Designing Protein Structures from Natural Language
Prompteina generates all-atom protein structures from natural language and experimentally yields nanomolar GM2A binders.
Prompteina adapts pretrained Proteina-Complexa so natural-language captions can condition all-atom protein structure generation. It is trained on Prompteina-Data, with 1.8 million structures and more than 50 million captions, and outperforms baselines on 15 of 18 structural properties. In binder design it leads on at least 13 of 20 targets and raises interface confidence 14 percent. Experimental tests against GM2A identified nine binders, the strongest with an apparent KD of 107 nM.
- Prompteina generates all-atom structures from natural-language design objectives.
- Training data contains 1.8 million structures and over 50 million captions.
- It leads baselines on 15 of 18 properties and at least 13 of 20 binder targets.
- Experiments identified nine GM2A binders, strongest at 107 nM apparent KD.
- AI models
- PrompteinaProteina-Complexa
From the original article
1 min read at arxiv.orgA central challenge in de novo protein design is generating proteins with desired structural and functional properties. Existing design models typically require a dedicated input representation for each property, limiting the design objectives they support.
By Joseph Lee · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.