1 week ago
Claude Produces Working Binders for 14 of 15 Drug Targets
Claude is an AI system made by Anthropic.
Researchers asked it to design tiny molecules called binders that attach to important proteins in the body.
The AI worked on 15 different targets and produced at least one successful binder for 14 of them.
Two outside companies made the designs and tested them in laboratories.
This matters because computer predictions often fail when scientists try to build them.
However, a binder is not yet a drug.
A possible medicine must also be safe, stable, selective and effective in people.
The findings were reported by Anthropic and have not yet been independently replicated by unaffiliated academic groups.
If the results hold up, AI could help researchers create and test drug candidates faster.
Anthropic says Claude generated working protein binders for 14 of 15 clinically significant targets.
Adaptyv Bio and Twist Bioscience synthesized and tested the designs without modifying them.
The targets included PD-L1, TREM2, TNF-alpha and EGFR.
Claude’s confirmed binder rates ranged from 22.6% to 35.1%, versus an estimated conventional rate of 10% to 15%.
The results demonstrate laboratory-tested binders, not medicines, and still require independent replication and extensive drug development.
- Who
- Anthropic’s Claude, with designs synthesized and tested by Adaptyv Bio and Twist Bioscience.
- What
- An autonomous AI campaign generated laboratory-confirmed protein binders for 14 of 15 targets.
- Where
- The designs were synthesized and tested in laboratories by Adaptyv Bio and Twist Bioscience; no location is specified.
- When
- The article does not specify when the campaigns or testing occurred.
- Why
- To evaluate whether an AI system could autonomously design working binders for clinically significant drug targets.
Case for Significance
Reasons for Caution
Laboratory validation
Case for Significance
The designs were physically synthesized and tested by Adaptyv Bio and Twist Bioscience, providing evidence beyond a computer benchmark.
Reasons for Caution
The testing was reported by the model’s developer and collaborators, and independent replication by unaffiliated academic groups has not yet occurred.
Performance against established targets
Case for Significance
Claude worked on difficult, clinically important targets and reportedly exceeded the estimated conventional de novo binder-design rate.
Reasons for Caution
The comparison depends on how the baseline is defined, and industry success rates vary by target class, method and reporting source.
Impact on drug discovery
Case for Significance
Higher hit rates could reduce the need for scarce specialist time and shift the bottleneck toward synthesis and testing capacity.
Reasons for Caution
Successful binding is only an early step; selectivity, stability, toxicity, manufacturing, delivery and human clinical trials remain necessary.
Key facts
- Targets tested
- 15 clinically significant protein targets
- Successful targets
- 14 targets produced at least one binder that worked
- AI system
- Claude, developed by Anthropic
- Confirmed binder rate
- 22.6% to 35.1%, depending on campaign structure
- Conventional baseline
- Approximately 10% to 15% for de novo binder design
- External testing
- Adaptyv Bio and Twist Bioscience synthesized and tested the submitted designs without modification
- Example targets
- PD-L1, TREM2, TNF-alpha and EGFR
- Development status
- The binders are laboratory results, not approved drugs
Sources
Related news
Novo Nordisk Taps AWS as Preferred Cloud and AI Partner
AI Helps Target ‘Undruggable’ Protein in Pancreatic Cancer
Indian Startup Uses Generative AI to Design Gene Editors
Indegene defies IT slowdown with 40% growth, pharma focus
UPSC GS-3 Practice Examines UPI Sustainability and AI Drug Discovery
Shamshad Ahmed Khan Advances Secure, AI-Driven Enterprise Transformation
How Workday Uses AI to Personalize Corporate Learning
Silver Lake weighs Workday takeover in historic software buyout
Google's Own AI Team Warns Its Hiring Filters Are Unreliable
