# Anthropic uses Claude to find and engineer bacterial and phage protein vaults > Anthropic researchers used Claude to map a phage-derived protein shell at 3.2-angstrom resolution, the first 3D structure of a vault particle from prokaryotic cells. The L05 shell encloses roughly 25% more volume than the human vault and can be resized by duplicating wall segments. - Type: AI news - Page: https://completeaitraining.com/news/anthropic-uses-claude-to-find-and-engineer-bacterial-and/ - Published: 2026-11-06 - Updated: 2026-11-05 - Author: Jeroen Erne - Jobs: Healthcare, Science and research - Reading time: 3 min Anthropic researchers used Claude to identify and produce vault-like protein particles from bacteria and phages in *E. coli*, reporting the results on November 4, 2026. The team mapped one phage-derived shell to a nominal resolution of 3.2 angstroms, establishing the first 3D structure of a vault particle from prokaryotic cells. The work gives scientists a faster-growing bacterial system for investigating these large, hollow protein shells, whose biological function remains unknown. #### From protein search to confirmed vaults The team used Claude Science to search a database of roughly 15 billion predicted proteins, finding major vault protein homologs in 29 bacterial phyla and four archaeal phyla. These proteins appeared in less than 1% of species with sequenced genomes, making them rare but broadly distributed. Researchers selected 44 proteins for cell-free expression testing before moving leading candidates into *E. coli*. Fifteen of 24 candidates produced soluble protein, and seven of those 15 formed vault-like particles under electron microscopy. Cryo-electron microscopy confirmed the vault shape in four cases - two from phages and two from bacteria. The phage-derived L05 particle was chosen for detailed structural analysis because it was among the best at forming complete, closed particles. #### Engineering the L05 shell L05 contains 96 copies of a single protein, arranged as 48 copies in each half. Mammalian vaults contain 78 copies. Each L05 protein has seven repeating wall segments, compared with nine in the human major vault protein. The resulting shell is shorter and wider than the human vault but encloses roughly 25% more volume. Its cap closes in three steps. Within each group of three neighboring protein copies, one ends at the first tier, another at the second, and the third continues to the tip. The copies are identical along the wall but adopt different conformations as the cap narrows. To identify contacts required for assembly, the team made 15 altered versions of L05. Some single substitutions near the wall seam or cap produced open, rosette-like particles or aggregates instead of closed vaults. Duplicating two wall segments gave L05 nine repeats and produced closed particles with approximately 20% more enclosed volume. The authors interpret this as evidence that repeat duplication can adjust vault size without preventing assembly, offering a possible route to size-tunable nanocapsules. Cargo encapsulation and delivery are proposed engineering directions, though the demonstrated result is control over shell geometry. #### Claude's role and the experimental limits The researchers selected the candidate proteins themselves. Claude converted the chosen sequences into DNA designs tailored to expression hosts, added gene-control sequences, and prepared the designs for ordering. An optimized size-exclusion chromatography method reduced the journey from cells to a microscopy grid to about 90 minutes, compared with roughly three days for the conventional sugar-gradient method developed for eukaryotic vaults. Claude initially argued against that approach because protein aggregates could emerge in the same excluded volume. It subsequently helped recommend a column, methods for tracking the protein and checking particle formation, and the direct loading of ruptured-cell contents without prior purification. For structural processing, Claude operated inside an internal multi-agent harness with computing resources, setting up and running the open-source RELION workflow while researchers guided analytical decisions. A first full model was ready after 34 hours, with final maps about a week after processing began. Anthropic estimates an initial model would otherwise have taken weeks. The project took under four months. The workflow also produced errors. One Claude-written protocol specified a growth-medium additive that interfered with introducing DNA into cells, delaying the project by two weeks. An early cryo-EM run skipped a standard step and generated a flawed map blended with its mirror image. Anthropic shared the laboratory notes with its model-training researchers and said subsequent Claude models appeared to make fewer DNA-design errors and identify experimental pitfalls more effectively. Each L05 protein copy contains a pocket with unresolved density from an unidentified small molecule. A computational search of 25,000 molecules identified coenzyme A or a derivative as a candidate ligand, though neither the molecule's identity nor any functional implication has been confirmed. The particles have only been produced in *E. coli*, and researchers have not tested whether they form in the bacteria and phages where the proteins naturally occur. Anthropic has submitted the technical report to bioRxiv and plans to release the structures and additional data later. #### Why this matters for science and research professionals The L05 system puts vault production into a fast-growing bacterial host with a 90-minute purification protocol, replacing a three-day eukaryotic method. Researchers now have a structurally mapped, resizable protein shell that can be modified through repeat duplication. The immediate experimental advance is control over geometry, not cargo delivery. For labs working on nanocapsules or protein engineering, the open structures and *E. coli* expression system lower the barrier to investigating vault assembly and modification. Professionals building AI-augmented lab workflows can also study the published error logs - the two-week delay from a protocol mistake and the flawed cryo-EM map - as concrete examples of where current AI assistance still requires human oversight. Those interested in applying AI to similar structural biology problems may want to explore [AI for Biochemists Training](https://completeaitraining.com/learningpath/ai-for-biochemists/) or browse [AI for Scientists Courses](https://completeaitraining.com/tag/science-and-research/) for relevant skill development. --- This is the Markdown form of https://completeaitraining.com/news/anthropic-uses-claude-to-find-and-engineer-bacterial-and/. The whole site is mapped at https://completeaitraining.com/llms.txt.