Claude Helps Solve the First Prokaryotic Vault Particle Structure

TL;DR: Anthropic’s biology research lab used Claude across sequence search, DNA construct design, purification optimization, and cryo-EM processing to solve the first 3D structure of a vault particle from a bacteriophage, a 96-copy shell of about 9.5 megadaltons. The work does not identify what vaults do, but it shows how AI can compress parts of fundamental biology research: vault purification fell from about three days to roughly 90 minutes, and a first structural model was ready 34 hours after processing began.

What Anthropic announced

Anthropic announced the first 3D structure of a vault particle found in prokaryotic cells—specifically, a vault built from a bacteriophage protein and produced inside a bacterium. Vaults are large, hollow protein shells found across the tree of life, and their biological function has remained unknown since their discovery in 1986. Related vault proteins had already been found in prokaryote and phage genomes, but Anthropic says no one had shown that they assemble into vault-like particles.

The finding is described in a preprint/technical report submitted to bioRxiv. Anthropic says it is releasing the findings early to show the scientific community how Claude can be used for discovery, and that it will release the structures and additional data soon.

Why vault particles are a longstanding mystery

Vaults were discovered in 1986 by Leonard Rome and Nancy Kedersha. They take their name from their shape: under an electron microscope, the ribbed shell resembled the vaulted ceiling of a cathedral. In mammals, the shell is made of 78 copies of the major vault protein, arranged as two identical halves joined at the middle. At roughly three times the mass of a ribosome, the vault is one of the largest known molecular complexes in eukaryotes.

Vaults are found across most of the eukaryotic tree, from slime molds to humans, which suggests they do something useful. Researchers have proposed that vaults transport molecules between the nucleus and cytoplasm, protect cells from toxic compounds such as cancer drugs, or support innate immunity. None of these proposals has been conclusively confirmed. Mice lacking vaults still survive, suggesting vaults may be necessary only under specific environmental conditions.

Vaults are also missing from organisms biologists rely on most for genetic experiments, including baker’s yeast, fruit flies, and nematode worms. In mice, animals bred without a vault partner protein or the vault RNA develop normally, leaving no obvious defect to follow up. A prokaryotic vault gives researchers a simpler system to study: bacteria grow quickly, making it easier to investigate where vaults came from, what their basic structure looks like, and what they contain.

The structure: 96 copies, 9.5 megadaltons, and a three-step cap

Anthropic used cryo-electron microscopy (cryo-EM) to determine the structure. The phage-derived vault is built from 96 copies—48 per half—of a single protein called L05, with a combined mass of about 9.5 megadaltons. The mammalian version has 39 copies per half. The phage vault is shorter and wider, roughly 54 nanometers long and 45 nanometers wide, compared with 67 by 40 nanometers in mammals. Despite being shorter, it encloses about 25% more volume because it is more spherical.

The protein chain that makes up the vault wall is built from seven repeating segments stacked end to end, rather than the nine found in mammalian vaults, and its end caps are flatter. Anthropic says it is among the largest protein shells ever identified in prokaryotes.

A notable feature is how the caps close. A wall 48 copies around must narrow toward the pole, and it does so in three steps, like a tiered dome. In every group of three neighboring copies, one ends at the first step, one at the second, and one continues to the tip. The three copies are identical along the wall but take different conformations in the cap.

Anthropic also engineered 15 altered versions of the protein, most with changes at the points where neighboring copies touch. Some single changes near the cap were enough to prevent the shell from closing: instead of sealed vaults, the team saw open, rosette-like particles, showing that this region is needed to seal the shell. When the team copied two of the seven repeating wall segments, giving the protein nine segments—the same number as in the human vault—it assembled into even larger vault particles, increasing the enclosed volume by approximately 20%. Anthropic says this demonstrates how vaults could have changed size over their evolutionary history by duplicating pieces of their own wall without disrupting assembly.

How Claude contributed

Claude Science searched a database of about 15 billion predicted proteins for relatives of the vault protein. The proteins turned out to be rare but widely distributed: present in 29 major groups (phyla) of bacteria and four of archaea, but in less than 1% of species with a sequenced genome. The team chose 44 proteins and tested each in a cell-free system. Claude turned each protein sequence into DNA tailored to the host cell, added the control sequences that switch a gene on, and prepared the designs for ordering. The best performers were moved into living E. coli bacteria, where 15 of 24 produced soluble protein.

For isolation, the conventional method—spinning the sample through a tube of increasingly dense sugar—takes approximately three days. The team tried size-exclusion chromatography instead. In this process, the sample flows through a column of porous beads; small molecules get stuck in the pores, while vaults are too big to enter and come out first in the void volume. Claude initially argued against the chromatography approach because clumps of misfolded protein also come out in that void volume. The team went ahead anyway, and Claude helped make it work by recommending a specific column and methods to track the vault protein and check whether particles had formed. Claude then proposed loading the contents of ruptured cells straight onto the column with no purification beforehand. The team was reluctant because the columns are expensive and raw E. coli contents can clog them or cause contamination, but the unconventional idea worked.

Anthropic notes that it is not the first group to isolate vaults on sizing columns, but its optimized method takes about 90 minutes from cells to microscopy grid instead of days. Under an electron microscope, seven of the 15 proteins had assembled into vault-like particles. Cryo-EM confirmed the vault shape for four of those seven, two from phages and two from bacteria. The team chose the phage protein L05 for the detailed structure because it was among the best at forming complete, closed particles.

Claude also processed the cryo-EM data. Working in an internal multi-agent harness with access to internal compute resources, Claude set up the open-source processing software RELION and ran each step while the team advised on analysis decisions. A first full model was ready in 34 hours, and the final maps were ready about a week after processing began. In under four months, the team screened dozens of candidate vaults, prepared samples for the microscope, and solved the first vault structure found in prokaryotes.

Anthropic says most of the work could be achieved by a typical lab with access to existing Claude products. Only the cryo-EM analysis, which used hundreds of GPUs in parallel to save time, required more computing resources than might be available to an academic institution.

What Claude got wrong

Anthropic also documented where Claude needed improvement. A protocol it wrote called for a growth medium containing an additive that interferes with getting DNA into cells, which set the project back by two weeks. Early in the cryo-EM processing, Claude skipped a standard step and produced a flawed map blended with its mirror image. The team told it to stop improvising:

“Please let’s do normal things unless they don’t work.”

The team shared its lab notes with the machine-learning researchers who train Claude so they could see where the model went wrong, and Anthropic says this informed the kind of data Claude learns from. Over the course of the project, Claude got better at these tasks as new models arrived, making seemingly fewer errors in DNA construct design and better identifying potential pitfalls in experiments. In areas the team knew well, it caught mistakes quickly, but mistakes cost time in areas the team did not know well. Overall, Anthropic says the benefits of working this way far outweighed the drawbacks, and Claude has become integral to its work.

An unexplained pocket and open questions

Claude pointed out a feature of the structure that the team cannot yet explain. A pocket partway up each of the 96 protein copies holds a small molecule that shows up clearly in the map but that the team has not been able to identify. Claude found that its shape resembles coenzyme A, a common helper molecule that cells use to shuttle chemical groups around. If it is coenzyme A, or something similar, it would be an early hint at a possible function.

Anthropic has so far made these vaults only in E. coli and has not tested whether they form in the bacteria and phages where they are naturally found. More work is needed to determine the function of these particles, and Anthropic says the work only deepens the mystery. Leonard Rome, whom Anthropic brought on to advise on the research, said:

“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.”

Rome also framed the open question this way:

“Will we find out that they are all doing the same thing? Or will we see an evolutionary development of vault function? Maybe they are doing the same thing as human vaults, and that is also very interesting.”

Anthropic says it will continue its research into the function of vaults and share more as the work progresses. The team also invites ideas about what these vaults might be doing, or systems in which to test them.