Researchers at Anthropic used the AI model Claude to create specific mini-proteins, known as binders, that target scorpion venom neurotoxins. According to a technical report, the model designed molecules in just a few days that physically bind to the toxins and could potentially form the basis of a new generation of antivenom.
More than 1.2 million scorpion stings are reported worldwide each year, causing up to 3,250 deaths, primarily among children with limited access to medical services. Current antivenoms are largely produced by immunizing horses and extracting antibodies from their blood. This process is time-consuming, expensive, and geographically limited, which is one reason scorpion envenoming is classified as a neglected tropical disease.
To address this problem and explore AI’s potential in drug discovery, the research team tasked Claude with designing binders against five different neurotoxins found in scorpions from Mexico, Brazil, North Africa, and the Middle East. These neurotoxins block sodium channels in human nerve cells, potentially causing heart failure and fluid accumulation in the lungs. The AI’s task was to design a molecule that could “catch” the toxin before it reached a nerve cell.
Within 64 hours of receiving the initial prompt, Claude worked with considerable autonomy: it used 12 different design tools, generated 318,484 protein designs, and selected the 180 best candidates for final testing through structural modeling. The research team subsequently added another 31 designs to the list in collaboration with Claude.
The 211 selected designs were tested under real laboratory conditions by partner organizations Adaptyv Bio and Twist Bioscience. Seventeen designs met stringent criteria, binding tightly to their target toxins. The two strongest binders identified in the study, Cn2-B1 and Cn2-B2, were designed to target Mexican scorpion venom and demonstrated binding affinity comparable to that of the best antibodies used to neutralize the venom in mice.
Despite the impressive laboratory results, the study’s authors emphasize that this is only the first stage. Before a mini-protein can become a real medicine, it must be tested in living cells (in vitro) and animals (in vivo) to establish whether it can actually neutralize the venom. In addition, the treatment must be safe, stable, and affordable for regions where scorpion stings are most common.
According to independent experts, binders designed by AI in days and validated in the laboratory within weeks point to a potential paradigm shift in the pharmaceutical industry. To help accelerate progress across the scientific community, the researchers have made all 211 designs and their laboratory data openly available.
