Anthropic and Citrini Research Test Claude Model for Scorpion Antivenom Design

What happened

Citrini Research and Anthropic collaborated to test whether an unreleased version of the Claude AI model could assist in drug development by designing proteins to bind to scorpion neurotoxins [1]. The experiment aimed to create protein binders that could attach to toxins, a necessary step for developing antivenom [2].

Key details of the process include:

• Anthropic provided access to an advanced, unreleased model after Citrini Research encountered biosecurity guardrails during initial research [1].
• Access was granted on the condition that the Citrini team worked from Anthropic’s offices [1].
• Anthropic covered compute and token costs, which were just under $50,000 [1].
• The team used a 35-page prompt and approximately 300 subagents to search for suitable proteins [1].
• Claude generated 318,484 protein designs [2].
• The team narrowed these down to 180 designs, then added 31 more with Claude’s help, totaling 211 designs [1, 2].
• The entire computational process took 64 hours [1].
• Testing was conducted by two contract research organizations, Swiss biotech company Adaptyv Bio and US-based Twist Biosciences, at a cost of approximately $150,000 [1].
• Out of 211 designs, 17 met the criteria for binding to one of four scorpion neurotoxins [2].
• None of the proteins bound to more than a single toxin [1].

Anthropic stated they have not confirmed if these binders neutralize venom in a cellular environment and have not produced an actual antivenom [2]. The results were based on recombinant toxins grown in a lab rather than purified venom [2].

Experts from the Technical University of Denmark, including professor Andreas Hougaard Laustsen-Kiel, assistant professor Melisa Benard Valle, and researcher Anne Ljungars, noted that while the results are technically promising, they are far from a medicine [1]. They stated that translating these binders into safe, stable, and effective treatments remains a long and challenging process [1]. Anthropic identified Cn2-B1 as one of the strongest candidates for potential future therapy [2].

Who is affected

This research impacts the fields of drug discovery and the life sciences. Anthropic is actively pursuing these areas as it approaches an anticipated IPO [1]. The findings are being released by Anthropic as a resource for the scientific field [2].