AI-based RNA structure prediction in 2026: overview
AI-based RNA structure prediction uses machine learning to infer how an RNA molecule folds from its sequence. The predicted structure can help explain how RNA performs regulatory, catalytic, or other biological functions, and can guide experiments when structural data are sparse. Models must contend with RNA’s flexibility and long-range interactions: small sequence changes or cellular conditions can alter the structures a molecule adopts.
Recent work is pushing beyond conventional structure prediction toward longer sequences and design. In October 2026, Xi'an Jiaotong University and Northwest Agriculture and Forestry University described UFold-X, a U-Mamba model targeting long-range secondary structure, while Shandong University presented the trRosettaRNA web server for 3D prediction. In September 2026, Stanford University and collaborators reported de novo design of RNA pseudoknots, showing how generative methods may connect structure prediction with RNA design. Limited structural data and generalization remain important constraints.
- 1Long-range challenge: Models targeting long RNAs address a persistent difficulty: distant interactions and limited training data can undermine reliable structure prediction.
- 2Prediction to design: De novo pseudoknot design signals a shift from predicting known folds toward generating RNA structures for intended functions.
- 3Physics still matters: Growing interest in physics-informed refinement reflects the need to pair learned patterns with biophysical constraints and experimentally grounded evaluation.
Top organizations working on AI-based RNA structure prediction
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 AI-based RNA structure prediction. Showing 7 of 116 organizations.
| # | Organization | Approach | Signals (24 mo) | Momentum | Why it matters |
|---|---|---|---|---|---|
| 1 | 
Harvard University
Research · United States |
Single-sequence language models | PUB 10 | +0% | Researchers affiliated with Harvard have developed a language model-based deep-learning approach for predicting RNA 3D structures from sequence. |
| 2 | 
Stanford University
Research · United States |
Generative structure models | PUB 5 · NEWS 7 | +0% | Recently, researchers affiliated with Stanford reported using deep learning to generate RNA pseudoknots de novo, demonstrating a route to designing intricate RNA structures. |
| 3 | 
Shandong University
Research · China |
MSA and co-evolution | PUB 7 · NEWS 1 | new | Recently, researchers affiliated with Shandong introduced the trRosettaRNA web server for automated RNA 3D structure prediction. |
| 4 | 
Chinese University of Hong Kong
Research · China |
Single-sequence language models | PUB 4 · NEWS 1 | -100% | Researchers affiliated with the Chinese University of Hong Kong have contributed to a language model-based deep-learning approach for RNA 3D structure prediction. |
| 5 | 
University of California
Research · United States |
Template-based modeling | PUB 8 | new | Researchers affiliated with the University of California have worked on template-based RNA structure prediction, including a blind code competition evaluating automated approaches. |
| 6 | 
Fudan University
Research · China |
Single-sequence language models | PUB 5 | -75% | Researchers affiliated with Fudan have contributed to language model-based RNA 3D prediction and developed an attention-and-ensemble method for RNA secondary structure prediction. |
| 7 | 
Griffith University
Research · Australia |
General | PUB 5 | -100% | Griffith researchers review deep-learning methods for RNA structure prediction, including transformer-based approaches and their potential use in RNA biology and drug design. |
| 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. |
Leading research centers
Leading research institutions 107 total
- Harvard UniversityUnited StatesResearchers affiliated with Harvard have developed a language model-based deep-learning approach for predicting RNA 3D structures from sequence.
- Stanford UniversityUnited StatesRecently, researchers affiliated with Stanford reported using deep learning to generate RNA pseudoknots de novo, demonstrating a route to designing intricate RNA structures.
- Shandong UniversityChinaRecently, researchers affiliated with Shandong introduced the trRosettaRNA web server for automated RNA 3D structure prediction.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
Where AI-based RNA structure prediction innovation happens
Share of documents with at least one organization from each country.
Open the interactive map with all 116 organizationsMain technical approaches
Last 24 months| Approach | What it is | Documents | Leading organizations |
|---|---|---|---|
| MSA and co-evolution | Uses patterns across related sequences to infer structural contacts; it can be informative when homologs exist but is less useful for poorly sampled RNAs. | 17 |
Purdue University, Pdb, Shandong University |
| Single-sequence language models | Learns sequence patterns from large RNA collections without requiring aligned homologs; performance can be limited by scarce structural labels and distribution shifts. | 23 |
Helmholtz Center Munich, Km 4, National University of Singapore |
| Graph and geometric learning | Represents nucleotides and their spatial relationships as graphs; it captures geometry directly but depends on effective representations and training data. | 29 |
University of Adelaide, Virginia Tech, University of Toronto |
| Generative structure models | Generates candidate RNA conformations or designs from learned distributions; it enables exploration but needs robust validation of physical plausibility. | 12 |
Helmholtz Center Munich, Virginia Tech, Bowling Green State University |
| Physics-informed prediction and refinement | Combines learned predictions with physical constraints or energy-based refinement; it can improve plausibility but may be computationally demanding. | 15 |
Helmholtz Center Munich, University of Toronto, Chittagong University of Engineering and Technology |
| Template-based modeling | Builds a structure using known RNA folds as references; it can work well for related structures but is constrained by template availability. | 12 |
Sun Yat Sen University, ShanghaiTech University, SYSU |
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
The reported funding data list $2,521,628 across 6 grants since 2020 and 6 funding-round records totaling $666,920,459. The round total includes broad university funding and should not be read as dedicated investment in RNA structure prediction.
Patent landscape
Top patent holders · last 24 monthsRecent developments
Latest news, duplicates removed- Oct 2026RNA structure and dynamics are crucial to its function. Despite decades of development in structural biology, the structural dynamics of RNA and its associated Mg2+ ions have never been directly visualized at…Nature - Issue - nature.com science feeds
- Jul 2026A pattern too common to be coincidence I first learned Science as a formal subject, when I was 11 years old, one of the first things I learned was, Bilateral and Radial symmetry in Nature. Now, it was not much…Science on Medium
Frequently asked questions about AI-based RNA structure prediction
What companies are working on AI-based RNA structure prediction?
FACTS list 9 companies overall and name Atomic Ai, Spidercore Inc., and Nanil Therapeutics Inc. among patent holders. These records indicate company-side intellectual-property activity, but they do not establish that each firm sells a prediction tool. Moderna Therapeutics also appears among organizations working on RNA-related applications.
Which AI-based RNA structure prediction startups are the most promising in 2026?
FACTS identify 0 startups in this field for 2026. The available information therefore does not support ranking promising startups. Company-side patent holders include Atomic Ai, Spidercore Inc., and Nanil Therapeutics Inc., but the data do not identify them as startups or provide evidence for comparing their products or commercial prospects.
Which universities and research centers lead AI-based RNA structure prediction research?
Harvard University, Stanford University, and Shandong University have 10, 12, and 8 documents in the 24-month window, respectively. Harvard researchers work on language-model prediction of RNA 3D structures; Stanford researchers have reported generative approaches; and Shandong University presented the trRosettaRNA prediction server in October 2026.
How much funding has AI-based RNA structure prediction received?
The reported figures are $2,521,628 across 6 grants since 2020 and $666,920,459 across 6 funding-round records. The latter total includes broad university funding, so it should not be treated as dedicated financing for RNA structure prediction. The available figures do not isolate a field-specific investment total.
Who holds the most AI-based RNA structure prediction patents, and is patent activity growing?
University of Maryland College Park, Spidercore Inc., Atomic Ai, and Nanil Therapeutics Inc. each have 1 listed patent, so no single holder leads in the supplied figures. There are 4 patents in the 24-month window, and reported patent growth is -100.0%. This indicates a decline in recorded activity under the stated comparison, not necessarily a fall in all patenting.
What are the latest developments in AI-based RNA structure prediction?
UFold-X, described in October 2026, targets long-range RNA secondary structure, while Shandong University’s trRosettaRNA offers web-based 3D prediction. In September 2026, Stanford University and collaborators reported de novo design of RNA pseudoknots. Together, these developments extend the field from fold prediction toward longer targets, accessible prediction tools, and structure generation.
What are the main approaches in AI-based RNA structure prediction, and how do they differ?
FACTS identify 6 approaches: MSA and co-evolution, single-sequence language models, graph and geometric learning, generative structure models, physics-informed prediction and refinement, and template-based modeling. They differ in their inputs and goals: some infer contacts from related sequences, while others learn sequence patterns, model geometry, generate candidates, apply physical constraints, or reuse known folds.
Which countries lead AI-based RNA structure prediction?
The United States leads with 256 documents, followed by China with 231 and the United Kingdom with 69. Japan and India follow with 54 and 49 documents, respectively. These figures describe recorded activity by country; they do not measure research quality or commercial readiness.
What are the applications of AI-based RNA structure prediction?
FACTS identify 5 application areas: RNA function and genome annotation, RNA therapeutic development, RNA-targeted drug discovery, viral and pathogen biology, and synthetic biology. RNA structure predictions can help researchers investigate molecular function and select candidates for further testing. The listed activity includes contributions from organizations such as Harvard University, Moderna Therapeutics, and Fudan University.