Anthropic and Citrini Research use an internal Claude model to autonomously generate protein minibinders against five scorpion neurotoxins, testing 211 designs

AI can now generate novel protein binders for toxins in days, bypassing previous bottlenecks in binding discovery.

Anthropic and Citrini Research have utilized an internal version of the Claude AI model to autonomously design protein minibinders capable of targeting specific scorpion neurotoxins. The AI was prompted to generate binders against five neurotoxins from four different scorpion species, focusing on the toxins' sites that interact with human sodium channels. Over a 64-hour period, Claude analyzed data, generated protein designs, and drafted a report, working with considerable autonomy. The AI produced 318,484 protein designs, from which 180 were selected for experimental testing. Collaborating with Adaptyv Bio and Twist Bioscience, researchers synthesized and tested 211 designs (including an additional 31 added with Claude's assistance). Of these, 17 designs demonstrated successful binding to their intended neurotoxin targets. This research aims to accelerate drug discovery for neglected diseases like scorpion envenomation, which affects millions annually with a high mortality rate, particularly in regions with limited medical resources.

This experiment signifies a potential paradigm shift in drug discovery, particularly for neglected diseases. By demonstrating that AI can autonomously design functional protein binders against complex targets like neurotoxins, it suggests that the rate-limiting step in developing therapeutics for conditions such as scorpion envenomation may shift from discovery to clinical translation and drug development. The accessibility of AI-driven design tools could empower smaller research groups or those with fewer resources to tackle challenging health problems previously deemed too complex or commercially unviable.

The ability to rapidly generate and test a large number of designs autonomously means that the discovery phase for new therapeutic molecules could be dramatically compressed. This speed, coupled with the potential for AI to optimize designs for specific properties, could bring much-needed treatments for neglected diseases closer to reality. The next critical step will be translating these identified binders into safe, affordable, and widely accessible drugs, a process that still requires significant investment and regulatory navigation.