Genomic and protein language models in 2026: overview
Genomic and protein language models learn patterns in DNA, RNA or protein sequences, creating representations that can support prediction and design. They matter because sequence data is abundant, while interpreting its biological function remains difficult. Models trained on evolutionary patterns can help prioritize variants, annotate proteins and guide protein engineering, but predictions still need careful validation against experiments and independent benchmarks.
Recent work is pushing beyond general-purpose sequence embeddings toward targeted prediction and broader design. In September 2026, UC Berkeley researchers’ GPN-Star was reported as a genomic model for identifying health-relevant variants. A September 2026 paper introduced LAMBDA, a benchmark for prophage detection, while Shanghai JiaoTong University researchers presented an enzyme-specific model for catalytic-property prediction. The direction is toward combining sequence with structural or evolutionary context, while testing whether models generalize to unfamiliar organisms and functions.
- 1From sequence to structure: Recent work combines sequence representations with structural cues to address tasks such as protein interaction and solubility prediction.
- 2Genome models expand: Genome-scale nucleotide models are gaining momentum, with activity up 75.3% across the comparison period in FACTS.
- 3Validation remains central: New benchmarks and task-specific models reflect a persistent gap between powerful learned representations and reliable biological predictions.
Top organizations working on genomic and protein language models
Ranked by Linknovate activity score · last 24 monthsThe activity score adds up the relevance of every patent, publication, grant and news item linking an organization to genomic and protein language models. Showing 7 of 1,791 organizations.
| # | Organization | Approach | Signals (24 mo) | Momentum | Why it matters |
|---|---|---|---|---|---|
| 1 | 
Stanford University
Research · United States |
Generative sequence models | PAT 2 · PUB 95 NEWS 147 |
+16% | Recently, Stanford researchers reported generative design of complete bacteriophage genomes using genome language models, extending sequence modeling from individual genes to whole viral genomes. |
| 2 | 
University of California
Research · United States |
Protein sequence models | PUB 104 · NEWS 14 | +18% | Recently, UC-affiliated researchers applied side-specific protein language models with geometric graph learning to predict how antibody–antigen mutations change binding free energy. |
| 3 | 
Massachusetts Institute of Technology
Research · United States |
Protein sequence models | PAT 4 · PUB 55 GRANTS 1 · NEWS 142 |
-8% | Recently, MIT researchers introduced MSA Pairformer to extend protein language modeling to protein–protein interactions, incorporating paired sequence information to capture coevolution. |
| 4 | 
Shanghai JiaoTong University
Research · China |
Protein sequence models | PUB 117 · NEWS 1 | -26% | Recently, researchers at Shanghai Jiao Tong University developed an enzyme-specific protein language model for predicting catalytic properties from amino acid sequences. |
| 5 | 
University of Electronic Science and Technology of China
Research · China |
Protein sequence models | PUB 105 | +2% | Researchers use protein language model representations in tools for predicting peptide activity, identifying CRISPR-associated proteins, and detecting protein modification sites. |
| 6 | 
Chinese Academy of Sciences CAS
Research · China |
Genome-scale nucleotide models | PUB 118 · NEWS 4 | +84% | Recently, researchers associated with CAS introduced LucaVirus, a unified nucleotide–protein language model for analyzing viral evolution and function. |
| 7 | 
Hainan University
Research · China |
Protein sequence models | PUB 61 · NEWS 11 | -20% | Researchers use protein language model representations in prediction tools for peptide activity, phosphorylation sites, and RNA–protein interactions. |
| 8 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
| 9 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
| 10 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
Startups to watch and leading research centers
Startups to watch 44 total
Cerebras SystemsUnited StatesCerebras partnered with Mayo Clinic on a genomic foundation model for predicting arthritis treatment response, combining genomic modeling with its AI computing platform.
Profluent Bio Inc.United StatesProfluent develops generative protein models for engineering proteins, including CRISPR enzymes with altered PAM specificity.
Integra TherapeuticsSpainRecently, Integra Therapeutics contributed to the discovery and language-model-guided design of hyperactive transposases for genome insertion.- ··Hidden startupCountryAvailable to registered users, with funding and founding year.
- ··Hidden startupCountryAvailable to registered users, with funding and founding year.
Leading research institutions 1,527 total
- Stanford UniversityUnited StatesRecently, Stanford researchers reported generative design of complete bacteriophage genomes using genome language models, extending sequence modeling from individual genes to whole viral genomes.
- Massachusetts Institute of TechnologyUnited StatesRecently, MIT researchers introduced MSA Pairformer to extend protein language modeling to protein–protein interactions, incorporating paired sequence information to capture coevolution.
- University of CaliforniaUnited StatesRecently, UC-affiliated researchers applied side-specific protein language models with geometric graph learning to predict how antibody–antigen mutations change binding free energy.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
Where genomic and protein language models innovation happens
Share of documents with at least one organization from each country.
Open the interactive map with all 1,791 organizationsMain technical approaches
Last 24 months| Approach | What it is | Documents | Leading organizations |
|---|---|---|---|
| Genome-scale nucleotide models | These models learn patterns across DNA or genomic sequences, enabling broad sequence analysis but facing challenges in long-context interpretation and biological validation. | 512 |
U.S. National Institutes of Health, University of Texas M. D. Anderson Cancer Center, Hainan University |
| Protein sequence models | These models learn representations from amino acid sequences for prediction and design, but sequence alone may not capture structural or interaction context. | 1,353 |
Massachusetts Institute of Technology, Duke University, Google DeepMind |
| RNA sequence models | These models learn RNA sequence patterns for functional prediction, with performance constrained by limited training data and RNA’s structural complexity. | 201 |
Shanghai JiaoTong University, Fudan University, Zhejiang University |
| Structure-integrated models | These models combine sequence representations with structural information, improving context for some tasks at the cost of added data and computational demands. | 480 |
University of Missouri, ESM, Google DeepMind |
| Generative sequence models | These models generate candidate biological sequences for design, but proposed sequences require experimental testing for function and safety. | 328 |
Chinese Academy of Sciences CAS, Stanford University, Shanghai JiaoTong University |
Emerging trends
Last 12 months vs previous 36 monthsTerms whose frequency in the last 12 months grew the most compared with the previous 36 months.
Funding and investment
Recent funding rounds
FACTS records $51,636,064 across 44 grants since 2020. These grant figures indicate research support, rather than a complete measure of commercial investment in the field.
Patent landscape
Top patent holders · last 24 monthsRecent developments
Latest news, duplicates removed- Sep 2026UC Berkeley researchers have created a new genomic language model, called GPN-Star, that excels at spotting genetic variants that impact human health. More than two decades after scientists first sequenced the entire…Scientific Inquirer
- Jul 2026The protein haemoglobin (artist’s illustration) can be miniaturized by an AI tool called Raygun.Credit: Evan Oto/SPL A sci-fi weapon has entered the AI age. A artificial-intelligence tool dubbed Raygun can miniaturize…Nature - Issue - nature.com science feeds
- Jul 2026Even though proteins evolve primarily residue by residue, natural selection does not evaluate them as such—it acts on holistic properties such as fold stability, binding affinity and catalytic activity, that emerge only…Nature - Issue - nature.com science feeds
- Jul 2026Jul 22, 2026 AI for Science: Breaking the boundaries of what science can achieve A new initiative aims to develop research-focused AI and build scientific infrastructure for faster discovery in the AI era. (Nanowerk…Nanowerk Nanotechnology and Emerging Technologies News
- Month 20XXHidden developmentAvailable to registered users with alerts enabled.
Frequently asked questions about genomic and protein language models
What companies are working on genomic and protein language models?
Google DeepMind and Nvidia are named corporate participants, while Cerebras Systems, Profluent Bio Inc. and Integra Therapeutics are listed as startups. Google DeepMind has applied deep learning to genome-scale sequence interpretation, and Nvidia provides accelerated computing for biological foundation models. The startups focus on genomic modeling, generative protein design and language-model-guided protein engineering.
Which genomic and protein language model startups are promising in 2026?
Cerebras Systems, Profluent Bio Inc. and Integra Therapeutics are the startups with specific activity described in FACTS. Cerebras partnered with Mayo Clinic on a genomic foundation model for treatment-response prediction. Profluent develops generative protein models, while Integra contributed to language-model-guided design of hyperactive transposases. Their listed work spans genome modeling and protein engineering.
Which universities and research centers lead research in this field?
Stanford University, Massachusetts Institute of Technology and University of California are among the research organizations with substantial activity in FACTS. Stanford researchers reported generative design of bacteriophage genomes; MIT introduced MSA Pairformer for protein interactions; and UC-affiliated researchers used protein language models to study antibody–antigen mutations. Chinese Academy of Sciences CAS is also active in genome-scale nucleotide models.
How much funding has genomic and protein language models received?
FACTS records $51,636,064 across 44 grants since 2020. This figure covers grants associated with the topic and does not establish the total amount of private investment or commercial funding. The dataset also lists 50 funding rounds, but its aggregate round value is not a reliable topic-specific estimate, so the grant total is the clearer funding indicator.
Who holds the most patents, and is patent activity growing?
Generation Bio Co. holds 7 listed patents and Sanofi S.A. holds 6; patent activity is down 48.0% in the comparison reported by FACTS. The topic has 152 patents in the defined window. These figures suggest that recent activity is more publication-led than patent-led, although patent counts alone do not measure the commercial value or technical importance of inventions.
How mature is genomic and protein language modeling?
The field is active but uneven in maturity: FACTS records 2,575 publications in the last year, up 46.2%, while patent activity is down 48.0% in its comparison. Protein models have broad research use, while genomic models still face questions about predictive performance and generalization. Recent benchmarks and task-specific models indicate that evaluation and validation remain important priorities.
What are the main approaches, and how do they differ?
FACTS groups the field into genome-scale nucleotide, protein sequence, RNA sequence, structure-integrated and generative sequence models. Nucleotide and RNA models learn from genetic sequences; protein models focus on amino acid sequences. Structure-integrated approaches add structural context, while generative models produce candidate sequences. Their trade-offs include data needs, computational cost, context captured and the need to test predictions experimentally.
Which countries lead genomic and protein language model research?
The United States leads with 47.2% of the documents in FACTS, followed by China at 32.1% and the United Kingdom at 15.8%. Germany and India are also among the countries with substantial activity. These figures describe the country distribution of documents, not the location of every research team or the share of commercial activity.
What are the applications of genomic and protein language models?
FACTS lists therapeutic discovery, protein annotation, variant interpretation, genome mining and synthetic biology, and RNA therapeutics as applications. Examples include predicting protein function and catalytic properties, interpreting genetic variants and mining genomes for useful biological systems. Generative models also support protein and sequence design, though candidates need experimental validation before practical use.