Claude Science pushes AI-designed protein binders
Anthropic reports that Claude autonomously designed functional protein binders for 14 of 15 targets, achieving 22.6%–35.1% experimental hit rates versus a typical 10%–15%. The work combines frontier models with specialist bioinformatics tools and independent wet-lab validation.
This is a meaningful step from AI-generated hypotheses toward experimentally validated scientific work, though binder design remains far short of complete drug discovery.
- –Claude handled target selection, computational design, tool use, and iteration with limited human intervention
- –Adaptyv Bio and Twist Bioscience independently built and tested the proposed proteins
- –The results suggest agentic AI can compress specialized protein-design workflows from months toward days
- –Strong binders are only an early proxy; toxicity, delivery, pharmacokinetics, and clinical efficacy remain unresolved
- –Claude Science’s broader significance is its tool-connected workflow, not just the headline hit rate
DISCOVERED
2h ago
2026-08-19
PUBLISHED
2h ago
2026-08-19
RELEVANCE
AUTHOR
Humpidumpi2