Claude Science proves autonomous protein binders with a 27% hit rate

Anthropic says its Claude Science platform can autonomously design protein binders and has achieved a 27% experimental hit rate for de novo protein binders across most targets tested. The company launched Claude Science on June 30, 2026. It integrates 60+ scientific databases and supports end-to-end protein design workflows, from target handling and candidate evaluation to results generation. Claude produces de novo designs using established computational biology pipelines such as RFdiffusion, ProteinMPNN, and AlphaFold. In Anthropic’s demonstration, about one in four designed binders showed genuine binding activity in wet-lab experiments. The article notes this compares competitively with reported experimental hit rates for similar AI protein design efforts (roughly 10% to 64% for mini-binders). Key caveat: peer-reviewed validation of Claude Science’s specific 27% results has not yet been publicly released, and independent replication may be needed before treating the figure as a benchmark. Anthropic frames Claude Science as an autonomy-first system, aiming for minimal human intervention while emphasizing reproducibility and traceability—features likely to matter for regulatory-sensitive pharmaceutical partners. Beta partner Manifold Bio reportedly used the platform to evaluate hundreds of binder candidates for tissue-targeting medicines.