Anthropic's biology team just mapped the first 3D structure of bacterial vault particles using Claude—potentially cracking a 38-year-old mystery.
Vault particles are massive barrel-shaped protein complexes (among the largest in cells) discovered in 1986, but their function is still unknown. Until now, vaults were only found in eukaryotic cells (animals, plants). This work proves bacterial proteins and bacteriophage proteins can also self-assemble into vault-like structures.
Why this matters technically: Finding vaults in simpler prokaryotic systems gives researchers a minimal model to reverse-engineer their function—way easier than working with complex eukaryotic machinery.
Claude's role: Accelerated structure determination from months of manual work down to days. However, the AI still made basic errors during the process, requiring collaboration with ML researchers to debug and iterate.
This is a real-world example of LLMs as scientific accelerators—not autonomous solvers, but powerful tools that compress tedious structural biology workflows when paired with expert oversight.
Vault particles are massive barrel-shaped protein complexes (among the largest in cells) discovered in 1986, but their function is still unknown. Until now, vaults were only found in eukaryotic cells (animals, plants). This work proves bacterial proteins and bacteriophage proteins can also self-assemble into vault-like structures.
Why this matters technically: Finding vaults in simpler prokaryotic systems gives researchers a minimal model to reverse-engineer their function—way easier than working with complex eukaryotic machinery.
Claude's role: Accelerated structure determination from months of manual work down to days. However, the AI still made basic errors during the process, requiring collaboration with ML researchers to debug and iterate.
This is a real-world example of LLMs as scientific accelerators—not autonomous solvers, but powerful tools that compress tedious structural biology workflows when paired with expert oversight.