Generate Biomedicines
Last reviewed
Jun 7, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 ยท 2,479 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 7, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 ยท 2,479 words
Add missing citations, update stale details, or suggest a clearer explanation.
Generate Biomedicines (styled "Generate:Biomedicines," and legally Generate Biomedicines, Inc.) is a clinical-stage biotechnology company that uses generative AI to design novel protein therapeutics, including antibodies and other engineered proteins. The company was founded in 2018 inside Flagship Pioneering, the Cambridge, Massachusetts venture creation firm behind Moderna, and is based in Somerville, Massachusetts. Generate built a computational and laboratory platform that treats protein design as a learnable, programmable problem, an approach the company calls "generative biology." Its best known research output is Chroma, a diffusion model for proteins published in Nature in November 2023. In February 2026 the company raised roughly $400 million in an initial public offering on the Nasdaq Global Select Market under the ticker GENB, one of the largest biotechnology IPOs in several years. As of mid-2026 it has multiple programs in clinical trials but no approved product.
Generate was conceived inside Flagship Pioneering's innovation foundry, Flagship Labs, where it began as two parallel explorations (internally designated FL56 and FL57) that were merged into a single company in 2019. The founding team comprised Flagship scientist-entrepreneurs Avak Kahvejian, Geoffrey von Maltzahn, and Molly Gibson, together with Gevorg Grigoryan, a former Dartmouth College computer science professor who became chief technology officer. Kahvejian and von Maltzahn served as co-founders, and Gibson took the role of chief innovation officer.
The premise was that the relationship between a protein's amino acid sequence, its three dimensional structure, and its biological function could be learned from large datasets and then run in reverse, generating new sequences predicted to fold into a desired shape and perform a desired job. This is closely related to the broader field of protein folding prediction that AlphaFold made famous, but inverted: rather than predicting structure from a known sequence, Generate set out to generate sequences for structures and functions that may not exist in nature.
The company operated in stealth until September 10, 2020, when it emerged publicly and described plans to design biologics against multiple disease targets, including the SARS-CoV-2 spike protein during the early COVID-19 pandemic. Mike Nally, a former Merck executive, joined as chief executive officer in March 2021 and has led the company through its clinical and public-market transitions.
Generate describes its core asset as the Generate Platform, which integrates three pillars: machine learning models, high throughput wet lab experimentation, and at scale structural determination (using methods such as cryo-electron microscopy and crystallography). The platform is designed as a closed loop in which AI proposes protein designs, the lab synthesizes and tests them, and the resulting experimental data feed back into the models. The company has said its models were trained on large corpora of biological data, on the order of roughly 160,000 protein structures and around 190 million genetic sequences.
The platform's most visible scientific result is Chroma. In a paper titled "Illuminating protein space with a programmable generative model," published in Nature on November 15, 2023, Generate researchers (with John Ingraham as lead author and Grigoryan among the senior authors) described a generative model that can directly sample novel protein structures and sequences. Chroma uses a diffusion process, the same general class of machine learning method behind image generators such as Stable Diffusion and DALL-E 2, but adapted to the geometry of protein backbones. It combines a correlated diffusion process over backbone atoms with graph neural networks for backbone synthesis and all-atom sequence design, and it scales roughly linearly with protein size, which lets it handle large complexes.
A distinguishing feature is conditioning: Chroma can be steered toward designs that satisfy external constraints such as symmetry, overall shape, a fixed substructure, or higher level semantic properties, including designs guided by natural language prompts about protein class. In the Nature study, the team experimentally characterized 310 proteins generated directly from the model without filtering, and reported that the great majority expressed, folded, and showed favorable biophysical properties, with several structures solved to confirm the intended folds. Grigoryan cautioned that "Chroma is a model and not a drug printer," noting that turning a designed protein into a therapeutic requires substantial additional work. Generate released the Chroma code as open source, with model weights made available to academic and non-profit researchers, which positioned the work alongside other open releases in AI for science. Chroma is one component of a broader internal toolkit; the company has indicated it uses additional machine learning models across antibody design and other modalities, and says the platform can produce designs for formats such as antibody-drug-conjugate payload neutralizers and engineered cell therapies in a matter of months.
Generate has advanced several internally designed proteins into the clinic, making it one of the first companies to test AI-designed antibodies in humans. Its lead program is GB-0895, an anti-TSLP antibody for severe asthma. The company's earlier and most cited program, GB-0669 against SARS-CoV-2, generated positive first-in-human data but was deprioritized in 2025 as commercial interest in COVID-19 antibodies faded.
| Program | Target / modality | Indication | Stage (as of 2026) | Notes |
|---|---|---|---|---|
| GB-0895 | Anti-TSLP monoclonal antibody | Severe asthma; COPD | Phase 3 (asthma); Phase 1b (COPD) | Lead asset; ~89 day half-life enabling 300 mg subcutaneous dosing every six months |
| GB-4362 | MMAE neutralizer | Antibody-drug-conjugate toxicity | Phase 1 (dosing expected mid-2026) | Received FDA Fast Track designation |
| GB-5267 | MUC16-directed armored CAR T cell therapy | Solid tumors (ovarian cancer focus) | Phase 1 (first patient expected H2 2026) | Developed with Roswell Park Comprehensive Cancer Center |
| GB-7624 | Anti-IL-13 monoclonal antibody | Atopic dermatitis | Phase 1 | First-in-human study in healthy adults |
| GB-0669 | Anti-SARS-CoV-2 neutralizing antibody (S2 stem helix) | COVID-19 | Deprioritized 2025 | Positive Phase 1 data; ~54 day half-life |
GB-0895 is the centerpiece. It is a long-acting monoclonal antibody that blocks thymic stromal lymphopoietin (TSLP), a cytokine that helps initiate airway inflammation; the same target is addressed by the marketed drug tezepelumab. Generate engineered GB-0895 for an extended half-life of roughly 89 days, which the company says supports a 300 mg subcutaneous dose given only twice a year. A Phase 1 study in about 96 mild-to-moderate asthma patients showed dose-proportional pharmacokinetics and sustained biomarker suppression across doses. On December 1, 2025, Generate announced the start of a global Phase 3 program, two replicate trials named SOLAIRIA-1 and SOLAIRIA-2, enrolling roughly 1,600 patients across more than 40 countries to measure asthma exacerbation reduction over 52 weeks. Generate has described this as the first global Phase 3 program for a long-acting anti-TSLP antibody and, by its account, the first Phase 3 study of an antibody designed using generative AI. Full enrollment was expected by around mid-2028.
GB-0669, the program that first put Generate on the map, is an antibody that binds a conserved epitope in the S2 stem helix of the SARS-CoV-2 spike protein, a region the company described as previously hard to target. Generate began a first-in-human trial in July 2022 and reported results on October 9, 2025, including a neutralizing index of 15 at the 1,200 mg dose (well above the threshold associated with efficacy), a half-life of about 54 days, and no dose-limiting toxicities or serious adverse events. Despite the data, the company said that "given the current commercial landscape, we have deprioritized internal development of GB-0669."
The remaining clinical programs reflect a shift toward immunology, inflammation, and oncology. GB-7624 is an anti-IL-13 antibody being studied for atopic dermatitis. GB-4362, a neutralizer of the cytotoxic payload MMAE that has received U.S. Food and Drug Administration Fast Track designation, is intended to manage antibody-drug-conjugate toxicity, with first dosing expected in mid-2026. GB-5267 is a MUC16-directed armored CAR T cell therapy for solid tumors, developed in collaboration with Roswell Park Comprehensive Cancer Center, with a first patient expected in the second half of 2026. The company also maintains preclinical programs against targets such as TL1A and OX40L and on combination antibodies. These efforts sit within the wider movement of AI drug discovery and AI in healthcare.
Two large pharmaceutical collaborations have validated and helped fund Generate's platform.
| Partner | Date announced | Upfront | Total potential value | Structure |
|---|---|---|---|---|
| Amgen | January 6, 2022 | $50 million | Up to ~$1.9 billion plus royalties | Initial five protein-therapeutic programs, option for five more; up to ~$370 million in milestones per program, royalties up to low double digits |
| Novartis | September 24, 2024 | $65 million (incl. $15 million equity) | Over $1 billion in milestones, plus tiered royalties | Multi-target discovery and development of protein therapeutics across undisclosed targets and disease areas |
Amgen signed the first deal on January 6, 2022, paying $50 million upfront to apply Generate's platform to an initial set of five clinical targets across multiple modalities, with an option to add up to five more programs. Amgen disclosed a potential transaction value of roughly $1.9 billion plus royalties, with up to about $370 million in milestones tied to each program, and it also agreed to invest in a future Generate financing round (which it did in the Series C). The Novartis collaboration, announced on September 24, 2024, gave Generate $65 million in upfront consideration, of which $15 million was an equity purchase, and made the company eligible for more than $1 billion in performance-based milestones plus tiered royalties up to low double digits. Generate CEO Mike Nally framed the Novartis deal as a way to "broaden the use of our cutting-edge generative biology platform," while Novartis biomedical research president Fiona Marshall described it as leveraging "the unique strengths of our respective companies." Combined, the Amgen and Novartis collaborations have generated on the order of $110 million in non-dilutive cash for Generate to date.
Before going public, Generate raised more than $800 million in private capital across three rounds, with Flagship Pioneering as the founding and lead investor.
| Round | Date | Amount | Notable investors |
|---|---|---|---|
| Series A | September 2020 | $50 million | Flagship Pioneering |
| Series B | November 18, 2021 | $370 million | Flagship Pioneering, ARCH Venture Partners, Fidelity, T. Rowe Price, Abu Dhabi Investment Authority subsidiary, Alaska Permanent Fund, Morningside Ventures |
| Series C | September 14, 2023 | $273 million | Amgen, NVIDIA (venture arm), plus prior backers |
| IPO (Nasdaq: GENB) | February 26, 2026 (priced) | 25 million shares at $16.00 |
The $370 million Series B in November 2021 was Generate's first external equity raise and drew sovereign wealth and crossover investors alongside ARCH Venture Partners, Fidelity, and T. Rowe Price. The September 2023 Series C added $273 million, with strategic participation from Amgen and the venture arm of NVIDIA, and was reported at the time as one of the larger biotech Series C rounds of the year.
On February 26, 2026, Generate priced its IPO of 25 million shares at $16.00 each for gross proceeds of about $400 million, and the stock began trading on the Nasdaq Global Select Market as GENB on February 27, 2026 (it opened below the offer price, at about $15). Net proceeds were roughly $369 million. Coverage described the offering as the sector's largest new biotech stock sale in nearly three years and part of the busiest month for biotech IPOs in over a year. In its first quarterly report as a public company, for the first quarter of 2026, Generate reported $7.2 million in revenue, a net loss of $61.7 million, and $516.6 million in cash, equivalents, and marketable securities as of March 31, 2026, which the company said was sufficient to fund operations into the first half of 2028.
Generate Biomedicines is one of the most prominent companies attempting to turn generative protein design from a research curiosity into approved medicines. Its publication of Chroma in Nature placed it among the small set of groups, alongside academic efforts such as RFdiffusion from the University of Washington, that demonstrated diffusion based de novo protein generation with experimental validation. The strategic interest of Amgen, Novartis, and NVIDIA, the move of multiple AI-designed antibodies into human trials, and a $400 million public offering together make Generate a closely watched test of whether AI-driven design can deliver a marketed drug. That test is not yet settled: the company has no approved product, its lead asset GB-0895 will not have pivotal data until later in the decade, and the deprioritization of GB-0669 despite positive clinical data shows that commercial and clinical risk remain even when the underlying technology works. The milestone and royalty figures attached to its partnerships are contingent on future progress and should not be read as realized revenue.