Anthropic: Claude helps map first vault particle from prokaryotic life

Anthropic posted on X that its biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life. The company published the full write-up on 4 November 2026 as Investigating vault particles with Claude. Vaults are barrel-shaped protein shells among the largest complexes in cells; their biological function is still unknown.

Anthropic’s opening on X said:

“We’re sharing how our biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life. Vault particles are a longstanding mystery. They are barrel-shaped objects, among the largest protein structures found in our cells. They were”

The remainder of that post points readers to the research page.

What Anthropic says it found

According to Anthropic, the structure comes from a bacteriophage protein the team calls L05. The particle is built from 96 copies of that protein (48 per half), versus 78 copies in human/mammalian vaults. Anthropic puts the phage vault at roughly 54 × 45 nm against about 67 × 40 nm for the mammalian form, with about 25% more enclosed volume despite being shorter, and a combined mass of about 9.5 megadaltons. The company says the map was solved by cryo-EM at roughly 3.2 Å resolution.

Anthropic frames this as the first 3D vault structure from prokaryotic/bacterial life. Related vault-like sequences had been spotted in prokaryote and phage genomes before, the post says, but — to Anthropic’s knowledge — no one had shown those proteins assemble into complete vaults. The company surveyed relatives across about 15 billion predicted proteins and reports hits in 29 bacterial and four archaeal phyla, though in less than 1% of sequenced species.

Function remains open. Anthropic says vaults made so far have only been produced in E. coli, and that a CoA-like pocket visible in the map is still unidentified — treat that as Anthropic’s structural hypothesis, not a confirmed ligand ID. The preprint (Athukoralage et al. 2026) has been submitted to bioRxiv; Anthropic says it will update the link when it goes live. The PDF Anthropic hosts is not peer-reviewed.

How Claude was used, per Anthropic

Anthropic says Claude helped search the protein space, design DNA constructs, speed size-exclusion isolation to about 90 minutes from a conventional ~3-day sugar-gradient prep, and drive RELION cryo-EM processing so a first full model was ready in about 34 hours. The company is clear that researchers directed the work and that Claude also made costly mistakes (a growth-medium additive that blocked transformation; an early map that mixed a structure with its mirror image).

Leonard Rome, who co-discovered eukaryotic vaults in 1986 and advised on the project, is quoted on Anthropic’s page:

“Figuring out the function of vaults is still enormously challenging. I am pleased that Anthropic has so much good data. Now, other people will be working on this problem … The broadness of Anthropic’s finding[s] suggests that it is going to be a complicated function and an interesting ride.”

Caveats

These claims come from Anthropic’s research post and X announcement, plus a company-hosted preprint PDF that has not yet been peer-reviewed. Independent labs have not, as of this desk’s check, published confirming structures. Anthropic itself says it has not yet tested whether the vaults form in the bacteria and phages where the genes naturally occur. Treat the prokaryotic vault structure as Anthropic’s reported result pending broader scientific review.

This report is not legal, financial, or health advice. See the disclaimer.