Adaptyv Bio and Anthropic launch a record AI protein design competition with over $2 million in support and a promise to publish even unsuccessful designs
Adaptyv Bio and Anthropic launch a record AI protein design competition with over $2 million in support and a promise to publish even unsuccessful designs
On 17 September 2026, Adaptyv Bio and Anthropic announced a partnership to launch an open AI protein design competition focused on potential drugs, with five challenges, each addressing a specific disease or biological mechanism. The first is already underway: participants must design a protein that binds to the EGFR receptor only in the acidic environment of a tumour, leaving healthy tissue unaffected. In some cancer cells, this receptor remains continuously active, driving their growth.
Participation is free, with three tracks available. Up to 30 teams from laboratories and companies will receive Claude credits worth up to $50 thousand for academic groups and $25 thousand for industry groups. Small teams and independent researchers will receive free access to the premium Claude Max 20x plan. The open track welcomes everyone, from enthusiasts to school students. The companies are jointly committing over $1 million to experimental validation, covering the synthesis and testing of more than 5000 proteins in Adaptyv Bio’s laboratory. Anthropic is separately providing over $1 million in Claude credits.
On the same day, Anthropic described the technical work behind the initiative: Claude independently wrote GPU kernels that accelerated a key operation in protein structure prediction by a factor of 1,7 to 3,2, while the full set of thirty open neural networks ran four times faster on average. In an official tweet, the company explicitly connected this work to the competition:
“To demonstrate what these speedups can achieve in practice, we are launching a protein design competition with Adaptyv Bio.”
Three weeks earlier, on 25 August, Adaptyv Bio raised $40 million to expand the same laboratory, which produces proteins from DNA templates without living cells and measures their binding to a target. Adaptyv Bio summed up the reason: “AI can move only as fast as experiments can keep up.” The new competition is the first major use of this increased capacity. Adaptyv Bio has already held four such competitions. For EGFR (2024, two rounds), the best binder’s affinity improved from 491 nM to 1,21 nM, with a lower value indicating stronger binding. For Nipah virus (2025 to 2026, with outbreak mortality reaching 75%), the value was 0,37 nM. For RBX1, a protein involved in cancer (early 2026), binding was weaker, at 23,7 nM. The current total of 5000+ designs is more than four times the size of the largest previous round, which tested 1200 designs in the Nipah competition.
The first challenge targets EGFR, a receptor already targeted by the approved antibodies cetuximab and panitumumab. These drugs act on the receptor throughout the body, including in healthy tissue, which contributes to their known tolerability problems. The new design must address this problem while meeting two requirements. It must bind to the target only in the acidic tumour environment (pH 6,5) and release it in healthy tissue (pH 7,4). It must also recognise the human and mouse versions of the receptor equally well, so that the same molecule can be tested first in mice and then in humans without redesign.
The companies will publish all results, including unsuccessful designs, under the open ODC-BY licence, which allows others to use the data freely with attribution. Participants retain the rights to their designs, and prizes will be non-cash only, with details to be announced later. Such failures are usually kept private in biotechnology. Publishing them in the open ProteinBase database gives other participants a chance to learn from them at no cost.