Research infrastructure · Protein engineering
Adaptyv Bio
A Swiss automated laboratory that turns AI designed protein sequences into measured binding data, and that has published the full results of three open design competitions, including the parts that made its own field look worse than advertised.
Thirty minutes and some kombucha
In December 2024, Adaptyv Bio published the results of its second open protein design competition. The task was to design a protein that binds EGFR, the receptor that Merck’s cancer antibody Cetuximab targets. A hundred and thirty designers submitted 1,131 sequences. Adaptyv synthesised and measured 400 of them.
The winner came from Cradle, a protein design company in the Netherlands. Its entry bound EGFR at 1.21 nanomolar, about eight times tighter than an scFv format of Cetuximab measured in the same run at 9.94 nanomolar. Four people produced it by pressing a few buttons and leaving a cluster running overnight.
Two details matter more than the headline. Cradle’s winning sequence had ranked between 300th and 600th on the computational leaderboard, because the scoring did not normalise for sequence length. And of the 400 designs that reached the instruments, only two beat the Cetuximab control at all. Adaptyv reported both.
The gap it sits in
Protein design got cheap in 2023 and 2024. RFdiffusion, BindCraft, ProteinMPNN and AlphaFold2 let a person with a laptop generate thousands of candidate binders in an afternoon. Testing them did not. A design still has to become DNA, then a real protein, then a measurement.
Julian Englert, the co-founder who runs the company, has used the same analogy since 2023. “Imagine that every time you used Github Copilot to generate some code, you had to wait 10 weeks for it to execute or to tell you that it had a bug,” he told Forbes. Adaptyv’s answer is a lab with no scientist in the loop: sequences arrive through a web platform or an API, the company orders the DNA, expresses the proteins in a cell free system rather than in living cells, and measures binding.
Four engineers from EPFL
Daniel Nakhaee-Zadeh Gutierrez, Amir Shahein, Moustafa Houmani and Julian Englert met at EPFL in Lausanne. All four are named as founders in EPFL’s August 2021 announcement, in the pre seed announcement of February 2022, and on the company’s SEC filing. The early money was small and local: CHF 140,000 in Swiss grants, USD 125,000 from Y Combinator, then a pre seed of about $2.5 million led by Wingman Ventures.
The first technology was not the one that shipped. Adaptyv set out to screen antibody binding on nanofluidic chips. According to an interview its community lead gave in 2026, that approach proved too experimental, and the team moved to established biophysical measurement in late 2023 before launching a platform in 2024.
Three competitions, and what they showed
Round one ran in September 2024. Of 726 submissions, 201 were made and tested. A hundred and forty six expressed, a 73% rate, and five bound, a hit rate of 2.5%. Round two, three months later, tested 400: 378 expressed and 53 bound. In the company’s words, “expression has basically been solved” in its cell free system.
Prediction was not. Adaptyv checked whether the computational scores it used to pick designs predicted the measurements. They did not. It stated flatly that “landing in the top 100 w.r.t in silico metrics played no detectable role in increasing the binding affinity,” and reported classifier performance near the coin flip line for AlphaFold2 interface PAE, ipTM and ESM2 pseudolikelihood.
The third competition picked a harder target: the Nipah virus attachment glycoprotein, a tetramer with little prior binder work, from a pathogen the WHO lists as a priority and for which no treatment exists. More than 680 participants submitted over 10,000 designs. Adaptyv tested 1,196. Expression fell to 86% and the hit rate to 9.3%, but 28 designs reached single digit nanomolar or better, and the best measured 0.37 nanomolar.
The de novo category was won by Nick Boyd of Escalante, who had stripped the filtering and ranking steps out of his pipeline and had not specified an epitope. Nine of his ten designs bound. The one he had hand curated was his worst.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| EGFR round 1 | Sep 2024. 726 submissions, 201 measured, 146 expressed (73%), five binders (2.5%), best KD 4.91e-7 M. Raw kinetic curves released under the Open Database License. | Company dataset |
| EGFR round 2 | Dec 2024. 130 designers, 1,131 submissions, 400 measured, 378 expressed (95%), 53 bound (13%). Best KD 1.21 nM against 9.94 nM for the Cetuximab scFv control. | Company dataset |
| Prediction versus measurement | The round 2 analysis found no significant correlation between AlphaFold2 ipTM, interface PAE or ESM2 pseudolikelihood and measured affinity, and ROC areas near 0.5 for classifying binders. Length normalised ESM2 pseudolikelihood reached 0.72. | Company dataset |
| Preprint | Both EGFR rounds written up on bioRxiv in April 2025, 27 authors, 601 designs characterised, optimised binders at 1.21 nM and de novo at 82 nM. Two authors disclosed as Adaptyv employees. No peer reviewed version found. | Preprint |
| Nipah competition | Jan 2026. 1,196 tested, 1,028 expressed (86%), 111 bound (9.3%), 28 at single digit nanomolar or better, best 3.7e-10 M. Neutralisation and serum albumin data released alongside. | Company dataset |
| Anthropic benchmark | Aug 2026. Adaptyv reports 1,320 blind designs, 95% expressed, 354 bound, a 26.8% hit rate. Anthropic reports 15 targets, binders against 14, hit rates of 22.6% to 35.1%, and a second testing lab. Adaptyv describes 16 targets with one excluded. | Company and partner |
| Assay method | The 2024 data packages describe Bio-Layer Interferometry. The 2026 Nipah release and Anthropic case study describe surface plasmon resonance. No bridging comparison published. | Company dataset |
| Corporate record | Adaptyv Biosystems Sàrl, Epalinges, entered in the Vaud register Nov 22, 2021, CHE-364.352.615, capital CHF 20,000, sole member Adaptyv Biosystems, Inc. of Delaware. | Public record |
| Founder count | 2021 and 2022 records name four founders and the SEC filing carries four signatures. Coverage of the 2026 round names two. The filing lists Nakhaee-Zadeh Gutierrez as COO; a 2025 report calls him CTO. | Differs from later accounts |
| Regulated products | No FDA 510(k) or PMA, no registered trial, no NIH or NSF award found as of Sep 22, 2026. | Not found |
Read plainly: the measured data is unusually good evidence, because Adaptyv released it and because the people whose designs failed wrote about it in public. The commercial picture is thinner. The customer count, the throughput multiple and the revenue growth are all company figures with no filed accounts behind them.
What to watch
- Whether the EGFR preprint reaches peer review, and whether the benchmark it proposes gets used outside the competition.
- A published comparison between the older Bio-Layer Interferometry results and the current surface plasmon resonance results.
- The London lab, and the plan to triple capacity by the end of 2026 and grow from 25 staff to about 60.
- Movement beyond binding into developability, peptides, degraders and cell based function, which the Series A was raised to fund.
In their words
“With an AI-designed protein, you can’t tell just from looking at it whether it will work in the way you want it to; it needs to be tested in a lab.”
Julian Englert, co-founder and CEO, Forbes, 2023 · Independent
“The proteins found in our bodies and in nature are the most advanced nanotechnology we know of.”
The four founders, pre seed announcement, 2022 · Company release
“Both in this competition and across the field, we still struggle to predict binding success through computational metrics.”
Adaptyv Bio, round 2 analysis, Dec 2024 · Company analysis
“In fact, a full 9 out of 10 of our designs bound for an astounding 90% success rate.”
Nick Boyd, Escalante, Jan 2026 · Competitor
“You were either in one of the bigger labs or you weren’t. And if you weren’t, you were designing stuff on the computer with no idea what it does in the real world.”
Tudor, community and protein engineering lead, Kiin Bio Weekly, 2026 · Interview
“Out of 1,320 designs, 354 bound their target, giving Claude an average hit rate of 26.8%.”
Adaptyv Bio, Anthropic case study, Aug 2026 · Company-stated
Related companies
Sources
- Public recordForm C, Adaptyv Biosystems Inc, CIK 0001894760
- Public recordCommercial register entry, Adaptyv Biosystems Sàrl
- Company datasetEGFR competition round 1 results repository
- Company datasetEGFR competition round 2 results repository
- Company analysisProtein Design Competition: Has binder design been solved?
- PreprintCrowdsourced Protein Design: Lessons From the Adaptyv EGFR Binder Competition
- Competitor8x improvement in EGFR binding affinity
- CompetitorWinning the de novo portion of the Adaptyv Nipah binder competition
- IndependentDemocratising the science of protein design with Adaptyv Bio
- IndependentEPFL spin-off secures funding from Y Combinator
- IndependentLausanne-based Adaptyv raises €34.35 million Series A
- InterviewAdaptyv: Closing the Loop on Protein Design
- IndependentClaude accelerates protein design and analytical chemistry
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
