Adaptyv Bio

Adaptyv Bio (legal name Adaptyv Biosystems Sàrl) is a Swiss company that runs an automated wet lab for testing computationally designed proteins. Customers, and increasingly AI agents, upload protein sequences through a web portal or an API; the lab synthesizes the DNA, expresses the proteins and measures properties such as binding affinity, expression and thermostability, then returns the results as structured data.^[1][2] The company is based at the Biopole Life Science Campus in Epalinges, Lausanne, and was founded by Julian Englert (CEO) and Daniel Nakhaee-Zadeh (CTO).^[1][3][4] It is widely known in the protein design community for free, open design competitions and for Proteinbase, a public database of experimentally tested designs. In 2026 it served, with Twist Bioscience, as an external validation lab for an Anthropic protein design campaign and co-sponsored a protein design competition with the company.^[5][6][7]

Overview

History

In the post that introduced the company on April 12, 2023, Englert wrote that the founders "teamed up in early 2021 to build better infrastructure & tools for protein designers," did Y Combinator in the summer of 2021, then raised a pre-seed round led by Wingman Ventures and built their technology at the Biopole campus. The stated plan was to build "a foundry," an integrated lab that lets protein engineers iterate on designs quickly, innovating across the stack from synthetic biology techniques to modular lab workcells and lab robotics.^[4] That year Adaptyv also released two open-source tools: ProteinFlow, a Python library for preparing protein structure data for deep learning, and Automancer, modular lab-automation software.^[12]

Adaptyv opened a public beta of its platform on June 13, 2024, pitching managed wet-lab infrastructure so that designers "can focus on designing, rather than pipetting," and comparing the model to cloud providers that freed software developers from running their own servers.^[13] A first version of its API followed in November 2024.^[12]

On September 18, 2025, the company dropped the beta label and opened the platform to anyone. It said more than 30 companies had begun using the lab that year, that it had tested "well over 10,000 proteins" in 2025, and that more than 10 preprints published that year used Adaptyv data. The same post disclosed an $8 million seed round led by Ace Ventures (which it described as raised 'earlier this year'; EU-Startups later dated the round to November 2024), with ByFounders and Founderful doubling down and LongGame and angel investors joining, and noted that the team had more than doubled in six months and moved into a new Lausanne lab and office.^[2]

On August 25, 2026, Adaptyv announced a $40 million Series A led by Highland Europe, with Ace Ventures, ByFounders and Y Combinator doubling down. It said lab throughput had grown more than fivefold over the previous year and that it had onboarded more than 100 customers, naming AI biology labs such as Chai Discovery and Boltz and pharmaceutical companies including Roche and Novo Nordisk. The company said it would double its Lausanne team and open a lab and office in London in the fourth quarter of 2026.^[11] EU-Startups, reporting the round as 34.35 million euros, added that Ace Ventures had led the seed round in November 2024 when Adaptyv had nine people, that the team had since tripled and revenue grown roughly tenfold, that DNA synthesis had been brought in-house and binding was measured by surface plasmon resonance, and that the company planned to roughly triple lab capacity by the end of 2026 and to grow its team from about 25 people to around 60.^[14]

The automated lab

Adaptyv describes its facility as an automated, AI-native wet lab. In its own account of the workflow, the platform converts a submitted amino-acid sequence into a DNA sequence, the DNA is assembled in the lab, and the protein is produced by cell-free protein synthesis, which runs the machinery a cell uses to read DNA and build proteins without a cell around it, handled by liquid-handling robots. Binding strength is then measured on dedicated instruments and processed into equilibrium dissociation constant (K_D) values.^[6]

On April 1, 2026, Adaptyv released its API generally, letting users and AI agents query the target catalogue, create experiments, track them through the pipeline, get cost estimates and pull structured results. The company said Tamarind Bio and Phylo had already integrated it.^[8] On May 6, 2026, Adaptyv became one of Benchling's Direct Ordering Partners, so that Benchling users can submit protein candidates for expression, binding or thermostability testing from a Benchling notebook or through Benchling AI, with prices shown inline before submission.^[9] Earlier case studies on the company's blog include wet-lab validation of Microsoft Research's EvoDiff designs and a maltose biosensor built with BindCraft.^[12]

Proteinbase

Adaptyv launched Proteinbase on October 6, 2025 as "a single hub for experimental protein design data," with more than 1,000 novel proteins at launch. Every entry comes from the Adaptyv lab under standardized protocols, links back to the design method that produced it, and includes negative results that are usually never shared. Submissions pass through a folding and annotation pipeline that assigns a predicted structure using Boltz-2, and the data is released under the ODC-BY license. Adaptyv offered a 20% discount on lab validation to customers who open-source their results there.^[10] The site organizes data by protein, collection, design method, target and designer profile, and it hosts the company's competitions.^[10][15]

As of September 2026 the Proteinbase competitions page reported 25,490 designs submitted across Adaptyv competitions by more than 1,000 designers in 42 countries, using 537 design methods, with 2,223 designs tested and 215 confirmed binders.^[15]

Protein design competitions

Adaptyv's competitions are free to enter and open to any design method. Adaptyv synthesizes the selected designs and tests them in the same lab under the same protocols so that results are directly comparable, and publishes every tested design with its sequence, predicted structure, design method and measurements, whether or not it worked.^[15] Adaptyv's own blog says the company was featured in a Nature news feature on protein design competitions, published on October 15, 2024 under the headline "AI has dreamt up a blizzard of new proteins. Do any of them actually work?"^[16][17]

Figures as listed on Proteinbase in September 2026.^[15] The RBX1 round was co-organized with the GEM Workshop at ICLR 2026.^[15]

EGFR rounds

The first two competitions asked designers to create binders against the extracellular domain of EGFR, a cancer-associated drug target. In the first round Adaptyv selected roughly 200 designs for screening, the top 100 ranked by AlphaFold2 interface pAE and the rest chosen across a range of design techniques; the winning combination was AlphaFold2 hallucination with SolubleMPNN inverse folding. Adaptyv noted that 398 of 726 submissions carried missing or unclear method descriptions and were excluded from its analysis.^[16]

In the second round, 378 of 400 tested proteins expressed (95%) and 53 bound EGFR, a 14% hit rate, against 2.5% and 76% expression three months earlier. Thirty of 130 participants produced at least one binder. Adaptyv attributed the jump to community knowledge sharing during the competition. The top-ranked entry by affinity came from Cradle, a variant of the approved antibody cetuximab carrying ten stabilizing framework mutations designed "in about 30 minutes"; second was a CDR-grafted nanobody from Chris Xu; third was Aurelia Bustos with a TGF-alpha scaffolding approach using RFdiffusion and ProteinMPNN. The best fully de novo design came from Lennart Nickel and Martin Pacesa using BindCraft.^[18]

Participants and organizers then wrote a community paper, "Crowdsourced Protein Design: Lessons From the Adaptyv EGFR Binder Competition," posted to bioRxiv in April 2025 with Tudor-Stefan Cotet and Igor Krawczuk as lead authors.^[19] Adaptyv's summary of the paper says the competition drew more than 1,857 submissions, of which 600 were tested, yielding 60 novel binders, and that the ranking metrics used during the rounds (ipAE, ipTM and ESM2 pseudo-likelihood) correlated only weakly with measured K_D on that dataset, while length-normalized ESM3 and ESMC correlated better. The paper closes by proposing BenchBB, the Bench-tested Binder Benchmark, a curated set of seven targets for wet-lab benchmarking of binder design.^[20]

Nipah and the TREM2 hackathon

The third competition asked for binders that disrupt the interaction between the Nipah virus glycoprotein G and its human receptor ephrin-B2/B3, a step the virus uses to enter host cells. Adaptyv reported more than 10,000 submitted designs, over five times the two previous competitions combined, from more than 650 participants, and selected 1,200 for testing, 600 of them by Boltz-2 ipSAE score and the remainder through expert review and a community vote.^[21][15]

In May 2026 Adaptyv published results from a one-day hackathon run with muni in San Francisco, in which 10 human teams and 6 fully autonomous LLM agents designed binders against TREM2, a target implicated in Alzheimer's disease. Of 141 designs, the top 100 by Boltz-2 ipSAE went to the lab and 37 bound. Human teams produced 25 binders from 65 designs (38.5%) and agents 12 from 35 (34.3%). The best human binder measured 1.11 nM and the best agent binder 3.64 nM, and all six agents independently picked the same design tool, which Adaptyv described as a hint of "tool monoculture."^[22]

Work with Anthropic

Validating Claude's binder designs

Adaptyv Bio and Twist Bioscience were the external evaluators for an Anthropic protein design campaign published on August 18, 2026, in which Claude Mythos Preview and Claude Opus 4.8, working inside Claude Science with open-source design and folding tools, designed binders against 16 targets, including all of Adaptyv's BenchBB set and two novel targets (15-PGDH and GDF-8) from its recent competitions. Anthropic reported 354 binders from 1,320 designs and success against 14 of the 15 targets for which usable data came back, with overall hit rates of 26.7% for Mythos Preview and 22.6% for Opus 4.8 in multi-target mode and 35.1% for Mythos Preview when it worked on one target at a time.^[7]

Adaptyv published its own account on August 20, 2026. It says the designs arrived anonymized, with the lab holding no information about which Claude model produced which protein, and that it ran affinity characterization by SPR at five target concentrations in duplicate. By Adaptyv's count 95% of the designs expressed and 354 of 1,320 bound, an overall hit rate of 26.8%, with one target (maltose binding protein) yielding no binders and one (mature GDF-8) excluded for low-quality measurements caused by target aggregation. Comparing against its own competitions, restricted to de novo minibinders, Adaptyv wrote that Claude's designs would have won five of six, reaching an 80% hit rate on TREM2 against the 38.3% baseline the case study gives for its TREM2 hackathon, improving the best 15-PGDH binder from 1.7 uM to 33.4 nM and the best RBX1 binder from 25.7 nM (the case study's competition figure; the Proteinbase listing gives 23.7 nM) to 3.9 nM, but failing to beat the best binder from the Nipah competition.^[6] Adaptyv characterized the campaign as an open-loop experiment and described closing the loop, with an agent learning from returned measurements and choosing its next batch, as the next step.^[6]

Anthropic x Adaptyv Protein Design Competition

On September 17, 2026, alongside research on accelerating open-source biomolecular models, Anthropic announced a protein design competition co-sponsored with Adaptyv. The two organizations jointly selected five problems described as being at the frontier of current capability, including species cross-reactivity, pH sensitivity and peptide-MHC specificity, as well as difficult targets such as GPCRs.^[5]

The two sponsors describe the funding differently and the distinction matters. Anthropic's post says that, together with Adaptyv and with contributions from Modal and Twist Bioscience, it is "committing up to $1 million in Claude credits and $250,000 in Modal compute credits, as well as wet lab validation for over 5,000 designs."^[5] The Proteinbase competition page instead says Anthropic and Adaptyv are "jointly sponsoring $1 million in experimental validation" to test more than 5,000 designs at no cost to participants, and that Anthropic is providing "an additional $1 million in Claude credits" to selected teams and researchers.^[15] Twist Bioscience provides DNA for the competition.^[5][15]

Schedule as published on the competition page, described there as planned and subject to change.^[15]

Entry is organized in three tracks. Track 1 is for research labs and companies with significant protein design experience: roughly 20 teams are selected, academic teams receive up to $50,000 in Claude credits and industry teams up to $25,000, and selected teams get a reserved allocation of up to about 18 designs per challenge. Track 2 is for affiliated individuals or teams of up to three, who receive complimentary Claude Max 20x access during the competition, with no guaranteed testing allocation. Track 3 is an open track for eligible participants outside the first two, self-supported and free to use any design method.^[15]

Anthropic and Adaptyv say designs will be selected for wet-lab testing through a Claude-based workflow that will be made public after the competition, that selection will not rely on a single in silico metric, and that participants retain ownership of their designs while results for tested designs, including negative ones, are published openly on Proteinbase. The competition has no cash prize, though the organizers say special non-cash prizes will be announced.^[15]

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Adaptyv Bio - Adaptyv Bio
2. ^1 ^2 ^3Adaptyv is the cloud lab for protein designers, and now available for everyone - Adaptyv Bio, September 18, 2025
3. ^1 ^2Team - Adaptyv Bio
4. ^1 ^2 ^3Introducing Adaptyv Bio - Adaptyv Bio, April 12, 2023
5. ^1 ^2 ^3 ^4How Claude is uplifting biomolecular modeling - Anthropic, September 17, 2026
6. ^1 ^2 ^3 ^4Case study: Benchmarking Claude's protein designs in the wet lab - Adaptyv Bio, August 20, 2026
7. ^1 ^2How Claude is accelerating protein design and analytical chemistry - Anthropic, August 18, 2026
8. ^1 ^2Adaptyv API: Giving everyone access to a real lab - Adaptyv Bio, April 1, 2026
9. ^1 ^2 ^3Test proteins with Adaptyv directly from Benchling - Adaptyv Bio, May 6, 2026
10. ^1 ^2 ^3Proteinbase: the home of protein design data - Adaptyv Bio, October 6, 2025
11. ^1 ^2$40M Series A to build the automated lab for agentic biology - Adaptyv Bio, August 25, 2026
12. ^1 ^2 ^3Blog - Adaptyv Bio
13. ^Generate lab data for your AI-designed proteins in just a few clicks - Adaptyv Bio, June 13, 2024
14. ^Lausanne-based Adaptyv raises 34.35 million euro Series A to build the automated lab for agentic biology - EU-Startups, August 27, 2026
15. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Anthropic x Adaptyv Protein Design Competition and Design Competitions - Proteinbase
16. ^1 ^2Protein Optimization 102: Lessons from the protein design competition - Adaptyv Bio, October 18, 2024
17. ^AI has dreamt up a blizzard of new proteins. Do any of them actually work? - Ewen Callaway, Nature, October 15, 2024
18. ^Protein Design Competition: Has binder design been solved? - Adaptyv Bio, December 25, 2024
19. ^Crowdsourced Protein Design: Lessons From the Adaptyv EGFR Binder Competition - bioRxiv, April 2025
20. ^Introducing BenchBB and the community paper of the Protein Design Competition - Adaptyv Bio, April 26, 2025
21. ^What happened in the Nipah Protein Design Competition so far? - Adaptyv Bio, December 9, 2025
22. ^Can LLMs design proteins? - Adaptyv Bio, May 13, 2026