The Information Machine
FollowingDay 4in the edition since 5 Nov 20262 sourcesupdated 5 Nov 2026

Anthropic's bacterial vault particle discovery

The gist

Latest: Anthropic Biology Team Maps First Bacterial Vault Particle Using Claude

The work extends the known distribution of vault particles to bacteria and bacteriophages, which could aid study of their function in human cells. It also serves as a concrete demonstration of both the potential and the limitations of using AI to accelerate scientific research.

When stories like this move, I write about them in an email each morning.

One free email each morning, around 8 AM US Eastern.

The full picture

Anthropic's biology research team used Claude to discover vault particles in bacteria and their viruses, producing the first 3D structural model of a bacterial vault particle. Vault particles are large barrel-shaped protein structures first discovered in 1986 by Leonard Rome and Nancy Kedersha; their biological function remains unknown. Until this work, vault particles had only been found in complex organisms like animals and plants. Claude ran the wet-lab pipeline end-to-end, cutting particle isolation from three days to about 90 minutes and delivering a first 3D structural model in 34 hours rather than the weeks to months typical of manual methods. The project also surfaced limitations: Claude made some simple mistakes that required collaboration with machine learning researchers to address.

How it developed
First in the edition · 5 November 2026
4 November 2026in the 5 Nov edition

Anthropic's biology team published results on November 4, 2026, showing that Claude identified vault particles in bacteria and bacteriophage proteins and produced the first 3D structural model of a bacterial vault particle.

Vault particles, barrel-shaped protein structures first discovered by Leonard Rome and Nancy Kedersha in 1986, had previously been found only in complex organisms like animals and plants, with their biological function still unknown. Claude ran the wet-lab pipeline end-to-end, cutting particle isolation from three days to about 90 minutes and delivering a structural model in 34 hours rather than the weeks to months typical of manual methods, though the team noted Claude made simple mistakes that required collaboration with machine learning researchers.

Sources

Get the daily email

When stories like this move, I write about them in an email each morning.

One free email each morning, around 8 AM US Eastern.

See the latest edition · Privacy