Citrini Research's attempt to create scorpion antivenom using AI faced significant hurdles, offering insights into AI's role in drug discovery.
Exploring AI's Role in Medicine
In a bold move to explore artificial intelligence's (AI) capabilities in drug development, Citrini Research partnered with Anthropic to determine if they could produce scorpion antivenom using an unreleased version of Claude, Anthropic's AI model. Despite their ambitious goals, the results fell short of expectations, highlighting both the potential and the limitations of AI in biopharmaceuticals.
The Background of the Initiative
Citrini, a firm known for its speculative research and prior fictional scenarios about economic crises linked to AI, openly acknowledged its lack of expertise in integrating AI with biological sciences. Nevertheless, they proceeded with the initiative and documented their journey in a blog post titled “We Wanted to Use Claude to Create a New Drug. Here’s What Happened.”
The objective was to leverage Claude to design proteins that would effectively bind to scorpion venom, a critical healthcare challenge in regions where stings can be fatal. This task is complicated by the high costs associated with scorpion antivenom production, which often involves extracting venom from horses.
Collaboration and Challenges Faced
To overcome initial setbacks with Claude's restrictions, Citrini sought collaboration with Anthropic, which agreed to facilitate access to a more advanced AI model. This partnership required Citrini to work on-site at Anthropic’s facilities, with the computational costs estimated to be around $50,000.
The team meticulously structured a 35-page prompt to guide Claude’s protein discovery, resulting in the AI generating approximately 320,000 candidates. After filtering the data, they narrowed it down to 211 promising protein designs.
Testing the Results
To validate these designs, Citrini enlisted external laboratories, particularly Swiss biotech Adaptyv Bio and US-based Twist Biosciences, committing around $150,000 to the evaluations. The outcomes revealed that while 17 of the proteins successfully bound to four of the targeted scorpion toxins, none were capable of binding to more than one toxin simultaneously. This limitation underscores the complexity of developing an effective treatment.
The Road Ahead: More Than Just Binding
Experts have pointed out that although discovering binding proteins is an important step, it is only part of a much larger process in drug development. The actual ability of these proteins to neutralize toxins is still untested, raising questions about their therapeutic viability.
Professors Melisa Benard Valle and Anne Ljungars from the Technical University of Denmark emphasized that while Citrini's findings hold technical promise, there remains significant uncertainty regarding the proteins' behavior within biological systems, their stability, and overall safety. They argue that the long and rigorous journey ahead involves turning promising candidates into functional medicines.
Reflecting on the Experiment
Despite the lack of a successful drug outcome, Citrini has framed this experiment as a valuable exploration of AI in the future of drug discovery. They aim to identify investment opportunities in the burgeoning field of AI-enabled healthcare. The company pointed towards potential beneficiaries of this technological trend, including firms specializing in drug testing.
The overall narrative emerging from Citrini’s experience suggests that while AI presents exciting possibilities for accelerating drug discovery, significant challenges remain in transforming its ideas into real-world solutions. This journey serves as a reminder that the path from AI-generated hypotheses to viable therapeutics involves intricate biological and technical hurdles that demand further research and investment.
