Generate Biomedicines: A Generative Biology Pioneer Programming the Future of Protein Therapeutics
Generate Biomedicines
New Drug Developer
Generate Biomedicines, Inc. was founded in 2018, incubated internally by Flagship Pioneering, the most prestigious entrepreneurial incubator in the global life sciences field, and is headquartered in Somerville, Massachusetts, USA. Unlike traditional venture capital, Flagship Pioneering adopts a unique "venture creation" model: its internal senior partners propose highly forward-looking "scientific hypotheses," which undergo years of confidential proof-of-concept validation in Flagship Labs. Once validated as feasible, they personally fund the establishment of the company, recruit executive teams, and bring it to market. Generate Biomedicines is another super star born from this industrialized "star-making assembly line," following Moderna.
The birth of the company can be traced back to 2016. At that time, "AI pharmaceuticals" mainly focused on using machine learning to screen known small-molecule compound libraries. However, two highly ambitious exploration projects — FL56 and FL57 — were initiated internally at Flagship. Gevorg Grigoryan (currently CTO of Generate Biomedicines), who was a professor of computer science and biochemistry at Dartmouth College at the time, made a groundbreaking discovery while participating in the projects: after conducting massive statistical analyses on the global Protein Data Bank (PDB), he proposed that the folding and arrangement of three-dimensional structures underlying proteins produced by nature over three billion years of evolution follow a kind of "universal grammar," similar to human language. This discovery formed the core of Generate Biomedicines' technological concept — if machine learning could master this "language of life," humans would be able to generate large protein molecules with specific therapeutic functions out of thin air, much like writing an essay or coding.
In 2018, based on this grand vision, Grigoryan, along with Noubar Afeyan, the founder and CEO of Flagship, and others, founded Generate:Biomedicines. Grigoryan holds a double bachelor's degree in biochemistry and computer science from the University of Maryland, Baltimore County, a Ph.D. in biology from the Massachusetts Institute of Technology, and has served as a professor at Dartmouth College, with a strong background in both computational biology and experimental biology.
Generate positions itself as a "Generative Biology" company, with its core mission being the shift from traditional "drug discovery" to "drug generation"—that is, instead of screening or modifying existing molecules found in nature, it leverages AI to "write" proteins with predefined therapeutic functions from scratch. The company has explicitly introduced the concept of "Programmable Biology," which implies the ability to design and encode biological functions with predetermined intent across therapeutic areas and protein modalities.
From the perspective of synthetic biology, Generate's technical system is highly aligned with the core paradigm of synthetic biology. The core engineering framework of synthetic biology is the "Design-Build-Test-Learn" (DBTL) cycle—a four-step iterative process aimed at optimizing biological systems to achieve specific, measurable performance goals. Generate Biomedicines' Generative Biology platform represents the cutting-edge practice of this paradigm in the field of protein drug development.
Specifically, the platform integrates the following key steps:
Generate: Utilize AI models (primarily Chroma) to design entirely new protein sequences and structures from scratch;
Build: Convert designs into real molecules through DNA assembly and protein production technology;
Test: Utilize high-throughput experimental platforms (including cryo-electron microscopy, multiplex bioassays, etc.) to validate protein functions and characteristics;
Learn: Feed experimental data back into the AI model to achieve continuous iterative optimization.
This closed-loop system enables the company to continuously evolve its design capabilities. Generate Biomedicines claims that the potential of its technology has been preliminarily validated, with three protein drugs developed through computational engineering now in human clinical trials.
Chroma is the core of Generate Biomedicines' technology ecosystem, with its research findings published in *Nature* in November 2023. According to the paper, Chroma is a generative model for proteins and protein complexes that can directly sample novel protein structures and sequences. It is also capable of steering the generation process toward desired characteristics and functions through conditional settings.
Technical Architecture: Chroma is built on the framework of diffusion models and graph neural networks. Diffusion models model high-dimensional distributions by gradually transforming them into simple distributions in a reversible manner; graph neural networks can effectively handle geometric information in complex molecular systems. In addition, Chroma introduces a diffusion process that respects the conformational statistics of polymer ensembles, an efficient neural architecture based on stochastic graph neural networks (supporting sub-quadratic scaling for long-range inference), and equivariant layers for efficiently synthesizing 3D structures from predicted inter-residue geometries. This technical system has led to Chroma being referred to as the "DALL-E" of the protein field in academic circles — capable of generating proteins with specific structures, functions, and even symmetries on demand.
Key Competencies:
Large-Scale Generation Capability: Generate Biomedicines has successfully designed and validated over 1 million proteins that have never existed in nature, which exhibited extremely high stability and functionality in preclinical tests.
Precise Condition Control: Chroma implements protein design as Bayesian inference under external constraints, which can involve symmetry, substructures, shape, semantics, or even natural language prompts. Scientists only need to input the parameters of the target, and Chroma can generate the corresponding protein structure.
Experimental Validation: Experimental characterization of 310 proteins shows that proteins sampled from Chroma can be expressed, folded, and possess favorable biophysical properties. The crystal structures of two designed proteins achieved atomic-level consistency with Chroma samples (backbone RMSD approximately 1.0 Å).
Open Source and Openness: Generate Biomedicines has open-sourced Chroma for use by academia and industry, demonstrating the company's strategic foresight in advancing the entire field.
Generate Biomedicines, Inc. centers its core competitiveness on an algorithm platform, establishing a dense patent layout around generative protein design. This encompasses general protein design rules as well as algorithms in specific application scenarios.
Core Algorithm Patent:
Diffusion Models for Protein Design (US20260038629A1): A technical system for de novo protein generation, including methods that receive protein feature targets and progressively adjust protein structures through a diffusion process.
Machine Learning-Guided Peptide Design (PCT/US2020/044646)
Optimization of Proteins Using Model-Based Methods (PCT/US2021/029197)
Deep Learning for De Novo Antibody Affinity Maturation and Property Improvement (PCT/US2021/043461)
Computer-Generated Binders (PCT/US2022/076970)
Using Machine Learning to Generate Cross-Reactive Neutralizing Antigen-Binding Molecules Against Viral Proteins (PCT/US2025/032314)
In addition, Generate Biomedicines is continuously expanding its patent portfolio and submitted a new patent application in August 2025 (Application No. 19300271). The company is also exploring the early integration of natural language with protein language models (US20250279161A1), aiming to develop intelligent systems capable of understanding and responding to multimodal queries for generating proteins for synthesis and small molecule drugs for manufacturing.
Generate Biomedicines has nearly 20 ongoing research projects, focusing on fields such as immunology, oncology, and infectious diseases. The following is an in-depth analysis based on the maturity of the pipeline:
5.1 GB-0895 (Anti-TSLP Long-Acting Antibody) —— The World's First AI-Designed Antibody to Enter Phase III
GB-0895 is Generate Biomedicines' flagship asset, targeting thymic stromal lymphopoietin (TSLP), an epithelial cell-derived cytokine that plays a central triggering role in the pathogenesis of type 2 inflammatory diseases such as asthma.
Technical Route: GB-0895 is not an antibody designed from scratch but rather an AI-driven optimization and screening of the marketed TSLP antibody Tezspire using the Chroma model. The initial goal of Generate Biomedicines was to design an antibody with five times the TSLP binding capacity of Tezspire and extend the dosing frequency from once a month to once every three months. Through AI optimization, Generate modified approximately 20% of the CDR regions (complementarity-determining regions, responsible for recognizing and binding to the target protein). The final screening results showed a 20-fold increase in binding affinity (reaching approximately 100 fM), laying the foundation for the possibility of dosing once every six months — twice the company’s original goal. Additionally, the company engineered the Fc region with YTE mutations, significantly enhancing the antibody's binding affinity to FcRn under acidic conditions, largely avoiding lysosomal degradation and effectively extending its half-life.
Clinical Progress:
Phase I Clinical Trial (Mild to Moderate Asthma): Involved 80-96 subjects with a single subcutaneous injection of different doses (10 mg to 1200 mg). In terms of safety, most adverse events were mild to moderate, and serious adverse events were assessed as unrelated to the trial drug. Pharmacokinetic data showed that the plasma concentration-time curves of all dose groups exhibited a parallel trend, with drug exposure proportional to the administered dose. Regarding efficacy, eosinophil counts in the 100mg and higher dose groups rapidly decreased to near zero or very low levels, indicating effective inhibition of the TSLP pathway, which was sustained for six months. FeNO (a marker of airway eosinophilic inflammation) also significantly decreased starting from the 300mg dose and remained reduced for six months. The half-life was approximately 89 days.
COPD Indication: Key biomarker reduction demonstrated in a Phase 1b extension trial in severe COPD patients, with PK characteristics consistent with earlier trials.
Phase III Clinical Trials: In December 2025, Generate Biomedicines announced the launch of two global Phase III clinical trials—SOLAIRIA-1 and SOLAIRIA-2, planning to enroll approximately 1,600 adult and adolescent patients with severe asthma. On January 26, 2026, the first patient completed dosing, with full enrollment expected by the first half of 2028. Approximately $300 million from the IPO proceeds will be allocated to these two Phase III trials.
Milestone Significance: GB-0895 is the world's first antibody drug entirely designed by generative AI and has advanced to Phase III clinical trials. It took only about four years from molecular design to Phase III clinical trials, significantly compressing the traditional drug development cycle.
5.2 GB-4362 (MMAE Neutralizing Antibody) —— Addressing ADC Toxicity Challenges
5.2 GB-4362 (MMAE Neutralizing Antibody) —— Addressing ADC Toxicity Pain Points
GB-4362 is an antibody-drug conjugate (ADC) auxiliary therapy designed to selectively bind and eliminate free monomethyl auristatin E (MMAE) circulating in the bloodstream. MMAE serves as the cytotoxic payload of various ADC drugs, and when released from ADCs, it can cause severe systemic toxicity. The design goal of GB-4362 is to reduce free MMAE levels without interfering with intact ADC activity, thereby improving the therapeutic index, which may support more sustained dosing of existing drugs and/or higher dose administration of novel drugs. The drug received IND approval in December 2025, obtained FDA Fast Track designation on January 23, 2026, and is expected to initiate a Phase I dose-escalation trial in 2026.
GB-4362 (MMAE Neutralizing Monoclonal Antibody) —— Addressing ADC Toxicity Pain Points
GB-5267 is an IL-18-modified CAR-T cell therapy targeting MUC16, focusing on the treatment of solid tumors, particularly ovarian cancer. Generate Biomedicines utilizes its AI platform to design best-in-class CAR-T cell therapies with enhanced stability, expression levels, and functionality, completing each design cycle in approximately 4-6 months, significantly shorter than the traditional timeline. The drug's IND was submitted and approved in early December 2025 and will be sponsored by Roswell Park Comprehensive Cancer Center, with Phase I clinical trials set to commence in 2026.
GB-7624 is an antibody targeting IL-13 for atopic dermatitis, with plans to initiate Phase I clinical trials in early 2025. In addition, Generate Biomedicines is advancing multi-target pipelines including TL1A, OX40L, as well as combinations of TSLP and IL-13, TL1A and IL-23, covering various immune-inflammatory indications such as respiratory, dermatology, and gastrointestinal conditions.
GB-7624 (anti-IL-13) and other immunology pipelines.
GB-0669 is the first AI-generated neutralizing antibody against SARS-CoV-2, which has achieved positive Phase I clinical results. This antibody overcomes the limitations of traditional biologics, successfully targeting previously "undruggable" sites with speed and precision, while maintaining efficacy against different variants and related coronaviruses. The relevant results were announced at IDWeek 2024.
November 2021: Completed a $370 million Series B financing round, with investors including Flagship Pioneering, etc.
September 2023: Completed $273 million Series C financing round, with NVIDIA as an investor
February 27, 2026: Listed on Nasdaq at $16 per share, issuing 25 million shares to raise $400 million, becoming one of the largest IPOs in the history of the AI pharmaceuticals sector. Goldman Sachs and Morgan Stanley served as lead underwriters, with Piper Sandler, Guggenheim Securities, and Cantor acting as co-managers. The stock code is "GENB".
The IPO price was set at the midpoint of the $15-$17 pricing range, with a market value of approximately $1.91 billion on the first day of trading. Noubar Afeyan, founder of Flagship, holds about 57.7% of the company's shares.
Amgen: In January 2022, a collaboration was reached to develop protein therapeutics for five clinical targets. The upfront payment was $50 million, with a potential total value of up to $1.9 billion.
Novartis: A multi-target collaboration was reached in September 2024 to develop potential first-in-class and best-in-class molecules using the Generate platform. The upfront payment for the collaboration was $65 million, with a potential total value exceeding $1 billion.
In addition, Generate has established manufacturing partnerships with CDMO companies such as Lonza and WuXi AppTec.
As of December 31, 2025, the company's cumulative net loss reached $676 million; losses for 2024 and 2025 were $181 million and $223 million respectively, with research and development expenses dominating. All revenue came from collaborations: $20.5 million in 2024 and $31.9 million in 2025. The company holds $221.5 million in cash and cash equivalents, which is expected to support operations for approximately 12 months. The business model has yet to achieve self-sufficiency and remains highly reliant on external financing.
At the beginning of 2020, almost no AI-designed drugs had entered human trials; by mid-2025, dozens of new drug candidates driven by AI will have entered clinical trials. GB-0895, developed by Generate Biomedicines, is the world's first AI-designed antibody to reach Phase III trials, marking a milestone in the industry’s evolution. Mike Nally, the company's CEO who previously worked at Merck for nearly 18 years, managed product lines with over $40 billion in revenue, and played a key role in the commercialization of Keytruda (K药), has further strengthened the company’s clinical development and commercialization capabilities after joining.
Generate Biomedicines is in the highly competitive AI drug discovery sector. According to PitchBook data, Xaira Therapeutics, Isomorphic Labs, and Generate Biomedicines all reached unicorn status after completing their first two rounds of institutional funding. Key competitors include Absci, Recursion, and Insilico Medicine. Additionally, tech giants such as Alphabet (through Isomorphic Labs) and NVIDIA are also entering the drug discovery field, further intensifying competition.
From the perspective of the discipline of synthetic biology, Generate's technological system can be understood as an advanced engineering practice of synthetic biology in the field of protein drug design:
AI-Driven Version of DBTL: The traditional DBTL cycle in synthetic biology largely relies on manual design and experimental validation, limiting iterative efficiency. Generate Biomedicines embeds machine learning into every step of the DBTL process, enabling automatic exploration of the design space and rapid convergence.
From "Remodeling" to "Creation": Traditional synthetic biology has mainly focused on the remodeling and reconstruction of natural biological systems (such as metabolic pathway engineering). The Chroma model from Generate Biomedicines, Inc. achieves "creation" from scratch — designing proteins that do not exist in nature, elevating the "creation" capability of synthetic biology to new heights.
"The Programming of 'Life Language'": The theory proposed by the company's CTO Grigoryan that "protein three-dimensional structure folding follows a universal grammar" essentially represents a renewed understanding of life coding, providing a more fundamental design framework for synthetic biology.
8.
Clinical Translation Risk: AI-designed molecules perform well in the laboratory, but the success or failure of Phase III clinical trials remains the ultimate test. The results of the SOLAIRIA trial for GB-0895 are expected to be read in the first half of 2028, marking a decisive moment for the company's value.
Financial Sustainability: Continuous huge losses and high cash burn rate require the company to continue relying on financing or commercialization revenue in the future. The $400 million raised from the IPO is expected to support approximately 12 months of operations, with clear subsequent financing needs.
Competitive Pressure: As the AI pharmaceuticals track rapidly heats up, competition from traditional pharmaceutical companies, tech giants, and other AI biotech firms is becoming increasingly fierce. Generate Biomedicines needs to continuously maintain its technological leadership in the field of generative protein design.
Regulatory Acceptance: Regulatory standards for reviewing AI-designed drugs are still in the process of being gradually established. Although the FDA has granted GB-4362 Fast Track designation, broader regulatory acceptance will still take time.
Cross-Modal Expansion: The company is advancing early fusion research on natural language and protein language, aiming to develop intelligent systems capable of accepting hybrid-modal queries and generating corresponding outputs. This will greatly expand the platform's applicable scenarios.
Deepening Technical Capabilities: Generate Biomedicines continues to innovate in the fields of CryoEM, multiplex bioassays, and protein manufacturing, with some advancements surpassing current industry or academic levels.
Pipeline Expansion: Beyond immunology and oncology, the company plans to extend its platform capabilities to a broader range of therapeutic areas, including infectious diseases and rare diseases.
Generate Biomedicines represents a new paradigm that deeply integrates AI, synthetic biology, and drug development. Its core value lies not only in advancing AI-designed drugs to Phase III clinical trials, a milestone in itself, but also in demonstrating a complete pathway for "programmable biology" moving from theory to practice—from understanding the "grammar of life" to "writing" protein codes, and then achieving continuous evolution through the DBTL closed loop. For the field of synthetic biology, Generate’s successful practice provides a highly convincing industry benchmark for AI-driven "design-build-test-learn" cycles.
Welcome to add the editor's WeChat to join the communication group.
When adding a friend, please note: Name + Company + Position
Companies in the fields of synthetic biology and biomanufacturing are also welcome to submit articles.
