- Claude was involved end-to-end: gene screening, purification optimization, cryo-EM data processing and structure modeling
- Purification dropped from roughly three days to 90 minutes; structure computation from weeks to 34 hours
- The vault is among the largest protein complexes in cells, and its prokaryotic form had gone unsolved for 40 years
AI CURATED HOT STORY · 11/5/2026
Anthropic's Claude First to Solve Bacterial Vault Structure, Cutting Weeks to Days
Anthropic's life sciences team has used the Claude model to determine the first three-dimensional structure of a bacterial vault particle, compressing a process that typically takes weeks to months into just days. The study also demonstrates for the first time that vault-like proteins in bacteria and bacteriophages can self-assemble into vault particles, providing a simpler model for studying the function of these mysterious proteins in human cells.
Technically, Claude was deeply involved in sequence search, DNA design, column chromatography suggestions, and structure modeling, reducing a key particle separation step from about 3 days to 90 minutes and producing an initial 3D model within 34 hours. The entire project was completed in under 4 months. The team adopted a division of labor where researchers retained decision-making authority while Claude executed tasks and offered suggestions, and they openly documented Claude's simple errors and the correction process, offering a reusable engineering paradigm for AI-assisted science.
This achievement has substantial implications for structural biology and AI for Science. It not only lowers the experimental barrier and time cost for vault research but also demonstrates the direct productivity of LLMs in wet-lab workflows, potentially encouraging more labs to adopt human-AI collaboration to accelerate scientific discovery.
Hourly heat curve
168 fully observed hoursThe line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.
Coverage timeline
Total 4 reports · Latest first- 新智元 (微信公众号)T2·71 pts
- 宝玉 (@dotey)T2·78 pts
- First demonstration that bacterial and phage vault relatives can assemble into vault particles, giving researchers a more tractable model for an enigmatic structure
- Claude contributed to sequence search, DNA design, column chromatography suggestions and structure modeling, cutting isolation from about three days to 90 minutes
- The division of labor, with researchers deciding and Claude executing and suggesting, is a reusable collaboration pattern for AI-assisted science
- Rohan Paul (@rohanpaul_ai)T2·86 pts
- Claude identified vault particles in bacteria and phages for the first time, expanding the known distribution of this mysterious protein structure.
- Direct AI control of wet-lab pipelines reduced key steps (isolation and modeling) by an order of magnitude in time cost.
- The study reveals correction mechanisms for model errors during human-AI collaboration, offering a reusable engineering paradigm for AI for Science.
- Anthropic AI (@AnthropicAI)T1·62 pts
- Claude compressed vault structure determination from weeks-to-months into days
- First evidence that bacteria and their viruses can assemble vaults, giving simpler models for function studies
- The team disclosed Claude's simple errors and the human-in-the-loop correction process