What happened
Claude generated 318,484 candidate minibinder sequences, filtered them to 180 designs, and, with human‑guided follow‑ups, added 31 more for a total of 211 designs that were synthesized and tested by contract research organizations. Seventeen of those designs met strict binding criteria (KD ≤ 1 µM) against individual scorpion neurotoxins, with the two strongest binders (Cn2‑B1 and Cn2‑B2) showing single‑digit nanomolar affinity.
Anthropic prompted its internal Claude Science model with a detailed design brief for minibinders targeting five scorpion neurotoxins (Cn2, Ts1, AaH I, AaH II, LqhIII). Over a 64‑hour campaign, Claude autonomously generated 318,484 protein sequences, applied 12 computational design tools, and filtered out designs resembling known proteins or toxins.
From this pool, Claude selected 180 candidates for synthesis. The research team added 31 additional designs, bringing the total to 211. Two contract research organizations—Adaptyv Bio and Twist Bioscience—produced the proteins using cell‑free and mammalian expression systems, respectively, and measured binding via surface plasmon resonance.
Binding was considered successful only if both labs observed a dissociation constant (KD) of 1 µM or tighter and no off‑target binding to a control protein. Seventeen designs satisfied these criteria, all against single toxins; none of the multitoxin designs met the threshold. The strongest binders, Cn2‑B1 and Cn2‑B2, achieved single‑digit nanomolar KD values and competed with a known neutralizing antibody, suggesting potential functional relevance.
Anthropic has released the full set of 211 sequences and the associated binding data publicly, inviting the scientific community to build on the results. The company notes that no in‑vitro neutralization or in‑vivo efficacy studies have yet been performed, and that further development would be required before any therapeutic claim could be made.
Source details: anthropic.com ↗
Why it matters
The study shows that a large‑language‑model‑driven workflow can accelerate early‑stage protein‑binder discovery for neglected tropical diseases, compressing a process that traditionally takes months into days. If the binders can be further validated for neutralization, the approach could lower the cost and time required to develop antivenoms, a critical need given the high mortality from scorpion stings in low‑resource regions. The open release of all 211 designs and associated data also provides a reusable resource for the broader scientific community, potentially spurring additional AI‑assisted drug‑discovery projects.
The workflow demonstrates that large‑language‑model‑guided protein design can produce experimentally validated binders at a scale and speed previously unattainable, addressing a key bottleneck in antivenom research.
By targeting neglected tropical diseases, the approach aligns with global health priorities where commercial incentives are weak, potentially enabling public‑sector or philanthropic investment to accelerate treatment development.
Open‑sourcing the designs lowers barriers for other labs, fostering collaborative validation and iteration, which could lead to faster translation of AI‑generated candidates into clinical candidates.
The study also highlights current limitations: multitoxin binder design remains challenging, and binding does not guarantee functional neutralization, underscoring the need for downstream biological testing.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Future work will need to confirm whether the identified binders neutralize scorpion venom in cellular assays and animal models. Success could lead to collaborations or licensing deals to develop affordable antivenoms. Observers should also watch for extensions of Claude’s design to other toxin families or neglected diseases, and for any policy or funding initiatives that support AI‑driven biomedical research.
Results from cellular neutralization assays and animal efficacy studies on the top binders, especially Cn2‑B1/B2.
Potential partnerships between Anthropic, biotech firms, or NGOs to move promising binders toward clinical development.
Expansion of Claude’s design to other venomous species or disease targets, which would test the generality of the approach.
Regulatory or funding responses to AI‑generated biomedical candidates, which could shape the ecosystem for future AI‑driven drug discovery.