AI's Attempt at Scorpion Antivenom: A Lesson in Limitations

Citrini Research recently embarked on an ambitious experiment to explore the potential of artificial intelligence (AI) in the field of drug development. Partnering with Anthropic, they aimed to see if an unreleased version of their AI model, Claude, could revolutionize the creation of a scorpion antivenom. However, the project encountered numerous challenges and ultimately fell short of its goals.

The Initial Idea

The premise behind Citrini's experiment was straightforward: harness the power of AI to design a new medication. In their enthusiasm, the firm acknowledged its limited expertise in biology, admitting, "none of us could claim to know anything about AI as it related to biology." Nevertheless, they pressed on, showcasing their adventurous spirit in a blog post titled, "We Wanted to Use Claude to Create a New Drug. Here’s What Happened.”

Collaboration with Anthropic

For this venture, Citrini sought the assistance of Anthropic, a company with a vested interest in the life sciences and AI advancements. Their collaboration came about when Citrini encountered obstacles while attempting to research drug candidates using Claude’s biosecurity protocols. After extensive outreach, they secured an agreement with Anthropic to access a more advanced AI model, provided they worked from Anthropic's facilities.

This partnership came at a cost, with estimated expenses for computing and token usage nearing $50,000. Ignoring the financial burden, Citrini focused on a specific objective: to design proteins that effectively target the venom of scorpions, which poses a significant health threat in many regions.

Scorpion Antivenom Challenge

With a focus on scorpion antivenom—known for its high production costs and complex handling involving venom extraction from horses—the team crafted a comprehensive 35-page prompt to guide Claude in the discovery process. The AI generated nearly 320,000 candidates, which the team whittled down to 211 after extensive filtering.

To test these proteins’ effectiveness, Citrini collaborated with the Swiss biotech firm Adaptyv Bio and Twist Biosciences, spending approximately $150,000 on evaluations designed to assess whether the newly designed proteins could bind to the targeted toxins. The results indicated that 17 of the proteins successfully interacted with four out of five scorpion toxins, but none managed to bind to multiple toxins, indicating significant limitations in their findings.

Long Way to Medicine

Despite the promising results of protein binding, Citrini’s experiment revealed a crucial gap: establishing that these proteins could neutralize the toxins had yet to be determined. As expert Andreas Hougaard Laustsen-Kiel pointed out, discovering effective binders is merely the first step; translating these findings into viable treatments poses a much larger challenge.

Academics Melisa Benard Valle and Anne Ljungars echoed this sentiment, cautioning that while the initial results were promising, they remained far from resulting in any functional medicine. There are many unknowns regarding the proteins’ safety, stability, and therapeutic potential that need comprehensive evaluation.

A Look Ahead

Despite not achieving its primary goal, Citrini has framed the experiment as a critical learning experience and a step toward understanding the investment landscape in AI-driven drug discovery. The firm advocates for further exploration of companies that might capitalize on this emerging trend, including those involved in testing and development.

The overarching lesson from this venture is clear: while AI shows great promise in accelerating the discovery process, the transition from concept to clinical application remains encumbered with complexities that require significant time and resources. While Citrini’s attempt may not have yielded a breakthrough drug, it signifies a growing interest in how AI can influence the future of healthcare and pharmaceuticals.