Anthropic uses Claude to map first 3D structure of a prokaryotic vault

The mystery of the vault

Vault particles are large, hollow protein shells found in nearly all cells across the tree of life. Although they were first discovered in 1986, scientists still do not know what they actually do. They are one of the largest known molecular complexes in eukaryotes, with a mass roughly three times that of a ribosome.

The name

Comparative vault geometry

The phage-derived vault discovered by the team differs significantly from the mammalian version in its physical dimensions. While the mammalian shell is longer, the prokaryotic version is notably shorter and wider. This specific shape allows the phage vault to enclose about 25% more volume because it is more spherical in its overall construction.

The protein chain forming the wall of this new vault is also distinct from its eukaryotic relatives. It is built from seven repeating segments stacked end to end, whereas mammalian vaults use nine. Additionally, the end caps of the phage vault are flatter, and the wall narrows toward the pole in three distinct steps like a tiered dome.

Searching the tree of life

To find these elusive proteins, the team used Claude Science to search a massive database containing about 15 billion predicted proteins. The search revealed that vault relatives are quite rare, appearing in less than 1% of species with a sequenced genome. However, they are widely distributed across the biological world, appearing in 29 major groups of bacteria and four groups of archaea.

After identifying these candidates, the researchers selected 44 specific proteins to move into the testing phase. They used a cell-free system to evaluate each of these proteins individually. The goal was to determine which specific genes would successfully yield a full-length protein and to measure the total amount of protein produced in each case.

AI-driven structural modeling

Mapping a biomolecular structure usually requires months of manual labor using specialized software. To bypass this, the team used an internal multi-agent harness that gave Claude access to internal compute resources. Claude was able to set up the open-source processing software known as RELION and execute each step of the workflow while the human researchers provided high-level advice on the analysis decisions.

The process was not without technical hurdles, as the model occasionally attempted to innovate in ways that produced errors. During the early stages of cryo-electron microscopy processing, Claude skipped a standard step and created a flawed map that was blended with its own mirror image. The researchers had to intervene and instruct the model to stick to standard procedures unless those methods failed to work.

Overcoming experimental hurdles

The research team faced significant obstacles because vaults are missing from the common organisms used for genetic experiments. Biologists typically rely on fruit flies, baker’s yeast, and nematode worms for these studies, but none of these species contain the vault protein. This absence is one of the primary reasons the particles have remained a mystery for forty years.

To overcome these limitations, the team looked toward prokaryotes and bacteriophages. They were willing to test an unconventional idea by loading the raw contents of ruptured cells directly onto a sizing column. While they were initially reluctant because these columns are expensive and prone to clogging, the experiment worked successfully and streamlined the purification process.

Engineering and future research

To understand how these proteins assemble, the team collaborated with Claude to create 15 different versions of the protein. Most of these alterations were made at the specific points where neighboring copies of the protein touch one another. These experiments helped reveal how the shell closes and how the vault might have changed in size over its evolutionary history by duplicating pieces of its own wall.

The team has submitted their findings to a preprint on bioRxiv and will share the structures and additional data soon. While the exact biological function of the vault remains unknown, the broadness of these findings suggests it will be a complicated and interesting problem for the scientific community to solve. Other researchers will now be able to use this new data to investigate the complex role these particles play in life.