What was announced
According to what was reported by IT之家, Anthropic's biology research team announced that it had determined for the first time the three-dimensional structure of a Vault, a vesicular protein complex of prokaryotic origin. The central finding concerns a protein of phage origin, named L05, capable of assembling into a complete particle made up of 96 identical chains, with a molecular mass of about 9.5 megadaltons and an internal volume about 25% larger than the analogous structure present in human cells. The stated resolution is 3.2 angstroms.
Vaults were first observed in 1986 and are among the largest cellular ribonucleoprotein complexes, present in about ten thousand copies per human cell and with typical dimensions of 67 by 40 nanometers. Although they have been known for about forty years, their biological function remains undetermined: Anthropic's work, as reported by the source, does not fill this gap and concerns structure, not physiological role.
How the research was conducted
The research began with a computational screen of approximately 15 billion predicted protein sequences, searching for homologs of the main Vault protein, MVP. Candidates were identified in 29 bacterial phyla and 4 archaeal phyla; 44 proteins were selected for laboratory testing. Of these, 7 expressed in Escherichia coli formed Vault-like particles, and cryo-electron microscopy confirmed the complete closed structure for 4 candidates, two bacterial and two phage.
Based on the L05 model, the team inserted additional wall repeat sequences, obtaining the L05+2 and L05+4 variants. The L05+2 variant shows an increase in both length and width and an internal volume a further 20% larger, bringing the total to about 50% more than the human Vault. The authors identify in this mechanism a possible engineering lever to modulate the size of the protein container. The analysis also detected an unassigned density on the inner wall of L05, compatible with coenzyme A or a derivative thereof.
Claude's role and its limitations
Claude was involved in several stages: protein search, DNA sequence design, optimization of the purification protocol, and processing of cryo-microscopy data. The source reports that a size-exclusion chromatography protocol reduced purification time from about three days to about 90 minutes, and that the first complete structural model was produced in 34 hours rather than in the weeks estimated for a conventional analysis.
However, the material also documents two problems attributed to AI. In one experiment, an additive suggested by Claude interfered with DNA delivery, causing a two-week delay; in processing the cryo-microscopy data, the system skipped standard steps, producing inaccurate structural maps, which were then corrected with closer supervision. These episodes, reported by the same source, should be considered as statements by the team and not as an independent assessment of the tool's reliability.
What remains to be verified
The structural data for L05 and the engineered variants have been deposited in the Protein Data Bank, and the manuscript is linked by the source. Several points nevertheless remain open: the function of Vaults in living cells has not been determined, the identity of the internal ligand is only hypothesized on the basis of computational research, and the potential use as a container for molecular transport is presented as a prospect, not as a demonstrated result. Overall, the case indicates a use of AI as an accelerator of technical stages with measurable benefits, but also the need for human verification throughout the process.
