# CRISPR

> Source: https://aiwiki.ai/wiki/crispr
> Updated: 2026-10-15
> Categories: AI for Science, Drug Discovery, Healthcare AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "CRISPR." aiwiki.ai, 15 Oct 2026. https://aiwiki.ai/wiki/crispr
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

CRISPR is a family of gene-editing tools derived from a bacterial immune system. The best known version, CRISPR-Cas9, uses a short guide RNA to direct the Cas9 nuclease to a matching DNA sequence, where the enzyme cuts both strands and the cell's own repair machinery produces an edit. Emmanuelle Charpentier and Jennifer Doudna shared the 2020 Nobel Prize in Chemistry "for the development of a method for genome editing," announced on October 7, 2020. [1] The first CRISPR therapy, Casgevy, was approved in the United Kingdom in November 2023, and in 2025 a single infant received a gene editor designed for his mutation alone.

CRISPR appears in an encyclopedia of artificial intelligence because almost every step of a modern editing experiment now runs through a learned model. Machine learning picks the guide RNA, predicts where the editor will cut by mistake, forecasts which repair outcome the cell will produce, and, since 2024, has produced working editors that do not exist in nature. The relationship runs the other way too: CRISPR screens generate the large perturbation datasets that biological [foundation models](https://aiwiki.ai/wiki/foundation_models) are trained on.

## How CRISPR editing works

In bacteria and archaea, clustered regularly interspaced short palindromic repeats are arrays of short repeats separated by spacers copied from the DNA of past viral infections. Transcribed into RNA, those spacers guide Cas nucleases to destroy matching sequences in a later infection, which makes the array a heritable record of what the cell has survived.

Engineered editing reuses that machinery. A single guide RNA of roughly 20 nucleotides is designed to match the intended genomic site, which must sit next to a short protospacer adjacent motif, or PAM. For the commonly used Cas9 from Streptococcus pyogenes (SpCas9), the PAM is NGG. The guide loads into Cas9, the complex scans the genome, unwinds matching DNA, and cuts both strands. What happens next is up to the cell. Non-homologous end joining usually produces small insertions or deletions that disrupt a gene; homology-directed repair, if a donor template is supplied, can write in a specified sequence, but it is far less efficient and works poorly in non-dividing cells.

Two problems shaped everything that followed. Guides differ enormously in how well they work, for reasons that are not obvious from the sequence. And a guide will often tolerate mismatches, cutting at unintended sites across the genome. Both problems are statistical pattern-recognition problems, which is why they became machine learning problems.

## Discovery and development

| Year | Development |
|---|---|
| 1987 | Yoshizumi Ishino and colleagues report unexplained repeated sequences near the iap gene of Escherichia coli [2] |
| 2000, 2005 | Francisco Mojica's group describes the repeat family across prokaryotes and proposes that the spacers come from foreign genetic elements [3][4] |
| 2002 | Ruud Jansen and colleagues coin the acronym CRISPR, by agreement with Mojica's group [5] |
| 2007 | Rodolphe Barrangou, Philippe Horvath and colleagues at Danisco show the system confers acquired antiviral resistance in Streptococcus thermophilus [6] |
| 2011 | Elitza Deltcheva, with Charpentier as senior author, identifies tracrRNA and its role in guide RNA maturation [7] |
| 2012 | Martin Jinek, Krzysztof Chylinski, Ines Fonfara, Michael Hauer, Doudna and Charpentier show Cas9 is a programmable dual-RNA-guided DNA endonuclease [8] |
| 2013 | Le Cong in Feng Zhang's laboratory and Prashant Mali in George Church's laboratory at [Harvard University](https://aiwiki.ai/wiki/harvard_university) independently demonstrate CRISPR editing in human cells [9][10] |
| 2016, 2017 | David Liu's laboratory reports cytosine and then adenine base editing [11][12] |
| 2019 | Andrew Anzalone and colleagues report prime editing [13] |
| 2020 | Charpentier and Doudna receive the Nobel Prize in Chemistry [1] |
| 2023 | Casgevy approved in the United Kingdom and the United States [14][15] |
| 2024 | Profluent releases OpenCRISPR-1, a gene editor generated by protein language models [16] |
| 2025 | A bespoke base-editing therapy is given to one infant with CPS1 deficiency [17] |

The intellectual property has been contested for more than a decade. In May 2025 the Court of Appeals for the Federal Circuit vacated a Patent Trial and Appeal Board determination on conception, holding that the board had conflated conception with reduction to practice, and remanded the case. On remand, in March 2026, the board again awarded priority for single-guide CRISPR-Cas9 in eukaryotic cells to the Broad Institute. [18][19]

## Base editing and prime editing

Cutting both DNA strands is a blunt instrument. Two later generations of editors avoid it.

Base editors fuse a catalytically impaired Cas9 to a deaminase enzyme, converting one base to another without a double-strand break. David Liu's laboratory reported cytosine base editing, which converts C to T, in Nature in 2016, and adenine base editing, converting A to G, in 2017. [11][12] Between them the two classes address a large fraction of known pathogenic point mutations.

Prime editing, reported by Anzalone and colleagues in 2019, goes further. A Cas9 nickase is fused to a reverse transcriptase and paired with an extended guide that carries the sequence to be written, so the editor can insert, delete or replace short stretches of DNA without a double-strand break or a separate donor template. [13] Prime Medicine announced the first human prime editing data in May 2025 for PM359 in p47-phox-deficient chronic granulomatous disease, reporting restored NADPH oxidase activity in 58 percent of neutrophils at day 15 and 66 percent at day 30 against a roughly 20 percent threshold associated with clinical benefit; the two-patient results were published in the New England Journal of Medicine. [20][21]

## Approved therapies

Casgevy (exagamglogene autotemcel), developed by Vertex Pharmaceuticals and CRISPR Therapeutics, became the first approved CRISPR medicine when the UK Medicines and Healthcare products Regulatory Agency authorized it on November 16, 2023. The US Food and Drug Administration approved it for sickle cell disease on December 8, 2023, with transfusion-dependent beta thalassemia following in January 2024. [14][15] It is an ex vivo therapy: a patient's blood stem cells are removed, edited to reactivate fetal hemoglobin, and returned after conditioning chemotherapy. The US list price is $2.2 million. In July 2026 the FDA expanded the label to patients aged 2 and older. [22] Uptake has been slow by pharmaceutical standards, with $115.8 million in 2025 revenue and roughly 301 patient initiations that year, though the pace rose through 2026. [23]

Several in vivo programs, which deliver the editor into the body rather than editing cells outside it, are in late-stage testing. Intellia Therapeutics reported that its Phase 3 HAELO trial of lonvoguran ziclumeran in hereditary angioedema cut mean monthly attacks by 87 percent against placebo; the FDA accepted the biologics license application with priority review in September 2026, with a target action date of March 10, 2027. [24] Approval would make it the first in vivo CRISPR therapy cleared anywhere. Intellia's ATTR amyloidosis program was interrupted when the FDA placed clinical holds on two Phase 3 trials in October 2025 after a serious liver safety event; the holds were lifted in January and March 2026 with added monitoring. [25] Beam Therapeutics has completed dosing in its base-editing sickle cell program, now called risto-cel, with results published in the New England Journal of Medicine in 2026, and reported that its in vivo base editor BEAM-302 raised alpha-1 antitrypsin above the protective threshold in all patients at the selected dose. [26][27] Eli Lilly acquired Verve Therapeutics in July 2025 for up to roughly $1.3 billion, taking over VERVE-102, an in vivo base editor targeting PCSK9 that reduced LDL cholesterol by a mean of 53 percent at its highest tested dose. [28]

## The KJ Muldoon case

In late February 2025 an infant known as KJ Muldoon, born with a severe urea cycle disorder caused by mutations in CPS1, received an in vivo base-editing therapy designed specifically for his variants. The work was led by Kiran Musunuru and Rebecca Ahrens-Nicklas at Children's Hospital of Philadelphia and the University of Pennsylvania, with contributions from the Innovative Genomics Institute, David Liu's laboratory, and several manufacturing partners. He received two infusions, at roughly seven and eight months of age. In the seven weeks after the first dose, his tolerance for dietary protein increased and his nitrogen-scavenger medication was halved, with no serious adverse events, including through intercurrent viral illnesses. The results were published in the New England Journal of Medicine on May 15, 2025, with the authors noting that longer follow-up is warranted. [17]

The case mattered beyond one patient because the therapy was designed, manufactured and administered roughly six months after his diagnosis as a newborn, a timeline no conventional drug development path supports. The FDA responded by proposing a regulatory route for individualized therapies: a November 2025 New England Journal of Medicine article by FDA officials described a "plausible mechanism pathway" using the case as its example, and the agency issued draft guidance in February 2026 covering individualized genome-editing and RNA therapies, allowing approval on one adequate and well-controlled investigation plus confirmatory evidence where randomized trials are not feasible. [29]

## Machine learning for guide design

The first predictors were not deep networks. Working at the Broad Institute, John Doench and colleagues measured the activity of thousands of guides and fit models to the result: Rule Set 1 in 2014, then Rule Set 2 in 2016, a gradient-boosted regression model developed with Nicolo Fusi and Jennifer Listgarten at [Microsoft Research](https://aiwiki.ai/wiki/microsoft_research). [30][31] Rule Set 2 was reimplemented as Azimuth and became the default scorer in the Broad's widely used guide designer.

[Deep learning](https://aiwiki.ai/wiki/deep_learning) followed. DeepCRISPR, published in Genome Biology in 2018 by Guohui Chuai and colleagues, pretrained a [neural network](https://aiwiki.ai/wiki/neural_network) on genome-wide guide sequences without labels, then fine-tuned it for on-target efficiency and off-target prediction in one framework, encoding each site as four nucleotide channels plus channels for epigenetic features. [32] DeepSpCas9, reported in Science Advances in 2019, was trained on a large single-library dataset and emphasized generalization to unseen cell types. [33] CRISPRon, published in Nature Communications in 2021, combined several public datasets with a new experimental set and improved efficiency prediction further. [34]

Progress since has been harder to assess than to produce. There is no authoritative head-to-head benchmark of on-target guide models from 2024 to 2026, and reviews have noted that newer deep learning tools have not been systematically evaluated against one another. Several recent transformer-based models exist only as preprints with self-reported benchmarks. The practical consequence is that guide design remains a field with many published models and little agreement about which to use.

## Predicting off-target activity

Off-target prediction is the harder half of the problem, because the thing being predicted is rare and the ground truth is expensive to collect.

The Cutting Frequency Determination score, published alongside Rule Set 2 in 2016, weights each mismatch between a guide and a candidate off-target site by position and identity. [31] Elevation, published in Nature Biomedical Engineering in January 2018 by Listgarten, Fusi and collaborators, replaced that hand-built score with two coupled machine learning models: one scoring individual guide-target pairs and one aggregating those scores into a single guide-level number, deployed as a cloud service alongside Azimuth. [35] CRISTA, published in PLOS Computational Biology in 2017, took a different angle, arguing that bulges as well as mismatches must be modeled and that structural and rigidity features around the PAM carry much of the predictive signal. [36] Deep learning versions followed, including the off-target branch of DeepCRISPR. Microsoft deployed Elevation and Azimuth together as a free guide-design service on Azure, with the code released openly. [37]

All of these depend on experimental measurements for training and validation. GUIDE-seq, reported in 2014, tags double-strand breaks in living cells with a short oligonucleotide. CIRCLE-seq, in 2017, moved the assay in vitro using circularized genomic DNA and requires no reference genome. CHANGE-seq, in 2020, used tagmentation to scale the approach, profiling 110 guide RNAs against roughly 200,000 off-target sites in primary human T cells. [38][39][40] Those datasets are the reason the models exist, and their composition is also the main source of bias in them.

## Predicting repair outcomes

For years the mutations produced after a Cas9 cut were treated as random. In 2018 three groups showed they are largely predictable. inDelphi, from Max Shen and colleagues working with David Liu, David Gifford and Richard Sherwood, trained a neural network and a nearest-neighbor model on a library of around 2,000 guides and predicted genotypes and frequencies of small insertions and deletions with a correlation of 0.87 across five human and mouse cell lines. The model also identified "precise-50" guides, between 5 and 11 percent of human-genome-targeting guides, for which a single outcome accounts for at least half of all major editing products, and the authors used them to correct pathogenic alleles in patient-derived fibroblasts. [41] FORECasT, reported the same month, reached similar conclusions with multinomial logistic regression, and Lindel followed in 2019. [42][43]

The same approach was extended to the newer editors. Deep learning models now predict base editing efficiencies and outcomes across many base editors, and the PRIDICT and PRIDICT2.0 models predict prime editing efficiency and product purity, the second version across different chromatin contexts. [44][45][46]

## AI-designed editors

The most direct intersection of CRISPR and artificial intelligence is the design of editors that nature never made.

[Profluent](https://aiwiki.ai/wiki/profluent) announced OpenCRISPR-1 on April 22, 2024, describing it as the first gene editor created by AI and released openly. The underlying work was published in Nature in 2025: the team curated more than one million CRISPR operons from 26 terabases of assembled genomes and metagenomes, trained [protein language model](https://aiwiki.ai/wiki/protein_language_model)s on them, and generated 4.8 times as many protein clusters across CRISPR-Cas families as are found in nature. Several generated editors showed comparable or improved activity and specificity relative to SpCas9 while sitting roughly 400 mutations away in sequence, and OpenCRISPR-1 also works in base-editing configurations. [16][47] Profluent reported a 63 percent hit rate across 209 tested Cas9-like proteins and reduced antibody reactivity compared with SpCas9, figures that come from the company rather than from the paper's abstract. The editors are licensed free for research and commercial use, including therapeutic use, under an agreement with ethical-use terms that prohibit human germline editing. Profluent open-sourced the editors, not the models that generated them. [48] The company announced its ProGen3 family of sparse protein language models, from 112 million to 46 billion parameters trained on 1.5 trillion tokens, in April 2025, and raised $106 million in November 2025. [49][50]

Two 2026 papers extended the idea. Doudna's laboratory reported structure and evolution-guided design of minimal RNA-guided nucleases in Science, coupling an inverse-folding model with evolution-informed residue constraints to generate divergent variants of TnpB, a compact Cas12-like nuclease. High-throughput screening produced editors that retained or exceeded wild-type activity in bacterial, plant and human cells, and cryo-electron microscopy of the most divergent active variant revealed new stabilizing contacts at the RNA and DNA interface. [51] A group led by Chengqi Yi at Peking University used [AlphaFold 3](https://aiwiki.ai/wiki/alphafold_3) differently, in a method called ContactSeek: rather than reading predicted structures, they used AlphaFold 3's predicted contact probabilities, correlated them with genome-wide off-target sequencing, and ranked consensus contact regions to find the residues that set specificity in Cas9-TadA adenine base editors. A variant combining one mutation in Cas9 and one in TadA8e outperformed several established high-fidelity base editors, and the method generalized to Cas12a-based cytosine base editors. [52]

## Genomic language models

Evo, published in Science in November 2024 by Eric Nguyen, Brian Hie, Patrick Hsu and colleagues at the [Arc Institute](https://aiwiki.ai/wiki/arc_institute) and [Stanford University](https://aiwiki.ai/wiki/stanford_university), is a 7-billion-parameter model trained at single-nucleotide resolution on 2.7 million prokaryotic and phage genomes with a context length of 131 kilobases. It was the first language model reported to co-design protein and RNA components together, and it generated functional CRISPR-Cas systems. Getting there was not effortless: Arc Institute reported that eleven designs were tested before one fully functional system, EvoCas9-1, worked, and it shares about 73 percent sequence identity with SpCas9. [53][54]

[Evo 2](https://aiwiki.ai/wiki/evo_2), released in February 2025 by the Arc Institute with [NVIDIA](https://aiwiki.ai/wiki/nvidia) and Stanford collaborators and published in Nature in 2026, scaled the approach to 7-billion and 40-billion parameter models trained on 9 trillion DNA base pairs across all domains of life, with a context window of one million tokens. [55] The earlier preprint gives the figure as 9.3 trillion DNA base pairs for the same corpus. [56] Its headline results are variant effect prediction, including on BRCA1, learned internal representations of exon-intron boundaries and transcription factor binding sites, and genome-scale generation; the CRISPR system generation result belongs to the first Evo, not to Evo 2. Weights, code and the OpenGenome2 dataset were released openly. [55][56]

Large language models have also been pointed at the experiment rather than the molecule. CRISPR-GPT, published in Nature Biomedical Engineering after appearing online in July 2025, is an agent that plans gene-editing experiments, selecting editing strategies, designing guides and drafting validation protocols. [57]

## Limits of the AI angle

No AI-designed gene editor has entered a human clinical trial. As of late 2026 there is no investigational new drug application, trial registration or clinical program using one, and the nearest commercial use is a cell-engineering service licensing OpenCRISPR-1 for stem cell editing. The gap between a designed editor that works in a cell line and one that can be manufactured, delivered, and shown to be safe in people is the same gap that limits [protein design](https://aiwiki.ai/wiki/protein_design) generally.

Guide and off-target models also transfer poorly. Predictors trained in one cell type, with one delivery method, at one expression level, routinely lose accuracy elsewhere, and the published benchmarks that exist are mostly self-reported by the authors of the model being measured. Several widely cited tools are no longer maintained. For therapeutic work, experimental off-target profiling remains mandatory rather than optional, which is a practical verdict on how far the models can be trusted.

## Ethics and governance

In November 2018 He Jiankui announced that he had edited the embryos of twin girls, known publicly as Lulu and Nana, to disable the CCR5 gene; a third edited child was later confirmed. The announcement came days before the Second International Summit on Human Genome Editing and was condemned almost universally. In December 2019 a court in Shenzhen sentenced He to three years in prison and a 3 million yuan fine for illegal medical practice, with shorter sentences for two collaborators. He was released in April 2022. Reports of his activities since have been inconsistent, including claims of a planned laboratory in the United States that have not been confirmed. [58][59]

Formal governance has remained advisory. The first International Summit on Human Genome Editing, held in Washington in December 2015, concluded that clinical use of germline editing would be irresponsible unless safety and efficacy issues were resolved and there was broad societal consensus. The second, in Hong Kong in November 2018, held that it "remains irresponsible at this time" while declining to call for a formal moratorium. [60] A 2019 Nature comment signed by Eric Lander, Francoise Baylis, Feng Zhang, Charpentier and others called for exactly that moratorium. [61] The World Health Organization's expert advisory committee, established in 2018, issued its recommendations and governance framework in July 2021, including a proposed confidential channel for reporting illegal or unsafe human genome editing research. [62] The third summit, at the Francis Crick Institute in March 2023, concluded that heritable human genome editing "remains unacceptable at this time" and added that the costs of current somatic gene therapies are unsustainable without a commitment to equitable access. [63] In May 2025 the International Society for Cell and Gene Therapy, the Alliance for Regenerative Medicine and the American Society of Gene and Cell Therapy jointly called for a ten-year global moratorium on heritable editing. [64]

That consensus is now being tested commercially. In October 2025 reporting revealed Preventive, a San Francisco public-benefit company that had raised around $30 million, the largest known investment in heritable editing, founded by Lucas Harrington, a former member of Doudna's laboratory, with investment from Sam Altman, Oliver Mulherin and Brian Armstrong. Harrington says the company is preclinical and will not proceed to human use if safety cannot be established; reporting indicates it has considered clinical work in the United Arab Emirates, where embryo editing is not prohibited. [65] In September 2026 NPR reported on Origin, a New York company led by Cathy Tie that tests experimental gene editors, some of them AI-designed, on donated embryos and cell lines with the stated aim of generating safety data to persuade Congress to lift the appropriations rider that bars the FDA from reviewing such applications. Researchers quoted in that coverage, including Robin Lovell-Badge of the Francis Crick Institute and Fyodor Urnov of the University of California, Berkeley, were critical, with Lovell-Badge saying the work offers false hope and Urnov questioning the companies' stated motives. [66]

## See also

- [Protein design](https://aiwiki.ai/wiki/protein_design)
- [Protein language model](https://aiwiki.ai/wiki/protein_language_model)
- [Evo 2 (genomic model)](https://aiwiki.ai/wiki/evo_2)
- [Profluent](https://aiwiki.ai/wiki/profluent)
- [AI in drug discovery](https://aiwiki.ai/wiki/ai_in_drug_discovery)
- [AI for Science](https://aiwiki.ai/wiki/ai_for_science)

## References

1. The Royal Swedish Academy of Sciences. (2020, October 7). Press release: The Nobel Prize in Chemistry 2020. https://www.nobelprize.org/prizes/chemistry/2020/press-release/
2. Ishino, Y., et al. (1987). Nucleotide sequence of the iap gene, responsible for alkaline phosphatase isozyme conversion in Escherichia coli, and identification of the gene product. Journal of Bacteriology, 169(12), 5429-5433. https://pubmed.ncbi.nlm.nih.gov/3316184/
3. Mojica, F. J. M., et al. (2000). Biological significance of a family of regularly spaced repeats in the genomes of Archaea, Bacteria and mitochondria. Molecular Microbiology, 36(1), 244-246. https://pubmed.ncbi.nlm.nih.gov/10760181/
4. Mojica, F. J. M., et al. (2005). Intervening sequences of regularly spaced prokaryotic repeats derive from foreign genetic elements. Journal of Molecular Evolution, 60(2), 174-182. https://pubmed.ncbi.nlm.nih.gov/15791728/
5. Jansen, R., van Embden, J. D. A., Gaastra, W., & Schouls, L. M. (2002). Identification of genes that are associated with DNA repeats in prokaryotes. Molecular Microbiology, 43(6), 1565-1575. https://pubmed.ncbi.nlm.nih.gov/11952905/
6. Barrangou, R., Fremaux, C., Deveau, H., Richards, M., Boyaval, P., Moineau, S., Romero, D. A., & Horvath, P. (2007). CRISPR provides acquired resistance against viruses in prokaryotes. Science, 315(5819), 1709-1712. https://doi.org/10.1126/science.1138140
7. Deltcheva, E., Chylinski, K., Sharma, C. M., et al. (2011). CRISPR RNA maturation by trans-encoded small RNA and host factor RNase III. Nature, 471(7340), 602-607. https://pubmed.ncbi.nlm.nih.gov/21455174/
8. Jinek, M., Chylinski, K., Fonfara, I., Hauer, M., Doudna, J. A., & Charpentier, E. (2012). A programmable dual-RNA-guided DNA endonuclease in adaptive bacterial immunity. Science, 337(6096), 816-821. https://pubmed.ncbi.nlm.nih.gov/22745249/
9. Cong, L., et al. (2013). Multiplex genome engineering using CRISPR/Cas systems. Science, 339(6121), 819-823. https://pubmed.ncbi.nlm.nih.gov/23287718/
10. Mali, P., et al. (2013). RNA-guided human genome engineering via Cas9. Science, 339(6121), 823-826. https://pubmed.ncbi.nlm.nih.gov/23287722/
11. Komor, A. C., Kim, Y. B., Packer, M. S., Zuris, J. A., & Liu, D. R. (2016). Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. Nature, 533(7603), 420-424. https://pubmed.ncbi.nlm.nih.gov/27096365/
12. Gaudelli, N. M., et al. (2017). Programmable base editing of A•T to G•C in genomic DNA without DNA cleavage. Nature, 551(7681), 464-471. https://pubmed.ncbi.nlm.nih.gov/29160308/
13. Anzalone, A. V., et al. (2019). Search-and-replace genome editing without double-strand breaks or donor DNA. Nature, 576(7785), 149-157. https://pubmed.ncbi.nlm.nih.gov/31634902/
14. Labiotech. Casgevy: the first CRISPR therapy approved in the UK. https://www.labiotech.eu/trends-news/casgevy-crispr-uk-approval/
15. Vertex Pharmaceuticals. (2023, December 8). Current report on the FDA approval of Casgevy. https://investors.vrtx.com/node/31276/html
16. BioSpace. (2024, April 22). Profluent successfully edits human genome with OpenCRISPR-1, the world's first AI-created and open-source gene editor. https://www.biospace.com/profluent-successfully-edits-human-genome-with-opencrispr-1-the-world-s-first-ai-created-and-open-source-gene-editor
17. Musunuru, K., Grandinette, S. A., et al. (2025). Patient-specific in vivo gene editing to treat a rare genetic disease. New England Journal of Medicine, 392(22), 2235-2243. https://doi.org/10.1056/NEJMoa2504747
18. Morrison & Foerster. (2025, May 13). Federal Circuit vacates and remands in the CRISPR interference. https://www.mofo.com/resources/insights/250513-federal-circuit-vacates-and-remands
19. Morrison & Foerster. (2026, April 7). The PTAB reaffirms priority decision for CRISPR IP. https://www.mofo.com/resources/insights/260407-the-ptab-reaffirms-priority-decision-for-crispr-ip
20. Prime Medicine. (2025, May 19). Prime Medicine announces breakthrough clinical data from the first patient treated with PM359. https://investors.primemedicine.com/news-releases/news-release-details/prime-medicine-announces-breakthrough-clinical-data-showing
21. Gori, J. L., et al. (2025). Prime editing for p47-phox-deficient chronic granulomatous disease. New England Journal of Medicine, 394, 1195-1203. https://pubmed.ncbi.nlm.nih.gov/41358590/
22. Healio. (2026, July 2). FDA expands Casgevy approval to children 2 years and older with sickle cell disease. https://www.healio.com/news/hematology-oncology/20260702/fda-expands-casgevy-approval-to-children-2-years-and-older-with-sickle-cell-disease
23. Precision Medicine Online. Vertex, CRISPR Therapeutics record $115.8M in 2025 Casgevy sales. https://www.precisionmedicineonline.com/business-news/vertex-crispr-therapeutics-record-1158m-2025-casgevy-sales
24. HCPLive. (2026, September). Lonvo-z BLA accepted with FDA priority review for hereditary angioedema. https://www.hcplive.com/view/lonvo-z-bla-accepted-fda-priority-review-hae
25. BioSpace. Intellia Therapeutics announces FDA lift of clinical hold on MAGNITUDE-2 Phase 3 clinical trial in ATTRv-PN. https://www.biospace.com/press-releases/intellia-therapeutics-announces-fda-lift-of-clinical-hold-on-magnitude-2-phase-3-clinical-trial-in-attrv-pn
26. Gupta, A. O., et al. (2026). Base editing of HBG1 and HBG2 promoters for sickle cell disease. New England Journal of Medicine, 394, 1824-1835. https://pubmed.ncbi.nlm.nih.gov/41931046/
27. Beam Therapeutics. (2026, August 4). Beam Therapeutics reports second quarter 2026 financial results. https://investors.beamtx.com/news-releases/news-release-details/beam-therapeutics-reports-second-quarter-2026-financial-results
28. Genetic Engineering & Biotechnology News. (2025, June). Lilly to acquire Verve Therapeutics for up to $1.3B. https://www.genengnews.com/topics/genome-editing/lilly-to-acquire-verve-therapeutics-for-up-to-1-3b
29. WTOP. (2026, February). FDA proposes new system for approving customized drugs and therapies for rare diseases. https://wtop.com/national/2026/02/fda-proposes-new-system-for-approving-customized-drugs-and-therapies-for-rare-diseases/
30. Doench, J. G., et al. (2014). Rational design of highly active sgRNAs for CRISPR-Cas9-mediated gene inactivation. Nature Biotechnology, 32(12), 1262-1267. https://pubmed.ncbi.nlm.nih.gov/25184501/
31. Doench, J. G., Fusi, N., Sullender, M., et al. (2016). Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9. Nature Biotechnology, 34(2), 184-191. https://doi.org/10.1038/nbt.3437
32. Chuai, G., et al. (2018). DeepCRISPR: optimized CRISPR guide RNA design by deep learning. Genome Biology, 19, 80. https://doi.org/10.1186/s13059-018-1459-4
33. Kim, H. K., et al. (2019). SpCas9 activity prediction by DeepSpCas9, a deep learning-based model with high generalization performance. Science Advances, 5(11), eaax9249. https://pubmed.ncbi.nlm.nih.gov/31723604/
34. Xiang, X., et al. (2021). Enhancing CRISPR-Cas9 gRNA efficiency prediction by data integration and deep learning. Nature Communications, 12, 3238. https://pubmed.ncbi.nlm.nih.gov/34050182/
35. Listgarten, J., Weinstein, M., Kleinstiver, B. P., et al. (2018). Prediction of off-target activities for the end-to-end design of CRISPR guide RNAs. Nature Biomedical Engineering, 2(1), 38-47. https://pubmed.ncbi.nlm.nih.gov/29998038/
36. Abadi, S., Yan, W. X., Amar, D., & Mayrose, I. (2017). A machine learning approach for predicting CRISPR-Cas9 cleavage efficiencies and patterns underlying its mechanism of action. PLOS Computational Biology, 13(10), e1005807. https://pubmed.ncbi.nlm.nih.gov/29036168/
37. Microsoft. CRISPR gene editing: Azimuth and Elevation as a cloud guide-design service. Microsoft AI Blog. https://blogs.microsoft.com/ai/crispr-gene-editing/
38. Tsai, S. Q., et al. (2015). GUIDE-seq enables genome-wide profiling of off-target cleavage by CRISPR-Cas nucleases. Nature Biotechnology, 33(2), 187-197. https://pubmed.ncbi.nlm.nih.gov/25513782/
39. Tsai, S. Q., et al. (2017). CIRCLE-seq: a highly sensitive in vitro screen for genome-wide CRISPR-Cas9 nuclease off-targets. Nature Methods, 14(6), 607-614. https://pubmed.ncbi.nlm.nih.gov/28459458/
40. Lazzarotto, C. R., et al. (2020). CHANGE-seq reveals genetic and epigenetic effects on CRISPR-Cas9 genome-wide activity. Nature Biotechnology, 38(11), 1317-1327. https://doi.org/10.1038/s41587-020-0555-7
41. Shen, M. W., et al. (2018). Predictable and precise template-free CRISPR editing of pathogenic variants. Nature, 563(7733), 646-651. https://pubmed.ncbi.nlm.nih.gov/30405244/
42. Allen, F., et al. (2018). Predicting the mutations generated by repair of Cas9-induced double-strand breaks. Nature Biotechnology, 37, 64-72. https://pubmed.ncbi.nlm.nih.gov/30480667/
43. Chen, W., et al. (2019). Massively parallel profiling and predictive modeling of the outcomes of CRISPR/Cas9-mediated double-strand break repair. Nucleic Acids Research, 47(15), 7989-8003. https://pubmed.ncbi.nlm.nih.gov/31165867/
44. Kim, N., et al. (2024). Deep learning models to predict the editing efficiencies and outcomes of diverse base editors. Nature Biotechnology, 42, 484-497. https://pubmed.ncbi.nlm.nih.gov/37188916/
45. Mathis, N., et al. (2023). Predicting prime editing efficiency and product purity by deep learning. Nature Biotechnology, 41, 1151-1159. https://pubmed.ncbi.nlm.nih.gov/36646933/
46. Mathis, N., et al. (2025). Machine learning prediction of prime editing efficiency across diverse chromatin contexts. Nature Biotechnology, 43, 712-719. https://pubmed.ncbi.nlm.nih.gov/38907037/
47. Ruffolo, J. A., Nayfach, S., Gallagher, J., et al. (2025). Design of highly functional genome editors by modelling CRISPR-Cas sequences. Nature, 645(8080), 518-525. https://doi.org/10.1038/s41586-025-09298-z
48. Mishcon de Reya. Profluent's OpenCRISPR-1: open sequence licensing in the life sciences industry. https://www.mishcon.com/news/profluents-opencrispr-1-open-sequence-licensing-in-the-life-sciences-industry
49. Profluent. ProGen3: scaling laws for protein design models. https://profluent.bio/showcase/progen3
50. SiliconANGLE. (2025, November 20). Profluent Bio raises $106M from Jeff Bezos and others to accelerate protein design AI. https://siliconangle.com/2025/11/20/profluent-bio-raises-106m-jeff-bezos-others-accelerate-protein-design-ai/
51. Skopintsev, P., Esain-Garcia, I., DeTurk, E. C., et al. (2026). Structure and evolution-guided design of minimal RNA-guided nucleases. Science, 393(6808), 313-318. https://doi.org/10.1126/science.aed6123
52. Meng, H., Lei, Z., Yan, Y., et al. (2026). Precise DNA base editing using AlphaFold3-based contact modelling. Nature, 656(8127), 463-473. https://doi.org/10.1038/s41586-026-10794-z
53. Nguyen, E., Poli, M., et al. (2024). Sequence modeling and design from molecular to genome scale with Evo. Science, 386(6723), eado9336. https://pubmed.ncbi.nlm.nih.gov/39541441/
54. Arc Institute. (2024, November 14). Evo: DNA foundation modeling from molecular to genome scale. https://arcinstitute.org/news/evo-science
55. Brixi, G., Durrant, M. G., Ku, J., et al. (2026). Genome modelling and design across all domains of life with Evo 2. Nature, 652(8112), 1349-1361. https://doi.org/10.1038/s41586-026-10176-5
56. Brixi, G., Durrant, M. G., Ku, J., et al. (2025, February 21). Genome modeling and design across all domains of life with Evo 2. bioRxiv. https://doi.org/10.1101/2025.02.18.638918
57. Qu, Y., et al. (2026). CRISPR-GPT for agentic automation of gene-editing experiments. Nature Biomedical Engineering, 10, 245-258. https://pubmed.ncbi.nlm.nih.gov/40738974/
58. The Scientist. (2019, December 30). China sentences gene-editing scientist to three years in jail. https://www.the-scientist.com/china-sentences-gene-editing-scientist-to-three-years-in-jail-66881
59. MIT Technology Review. (2022, April 4). He Jiankui, the scientist who gene-edited babies, is out of prison. https://www.technologyreview.com/2022/04/04/1048829/he-jiankui-prison-free-crispr-babies/
60. National Academies of Sciences, Engineering, and Medicine. (2018, November 28). Statement by the organizing committee of the Second International Summit on Human Genome Editing. https://www.nationalacademies.org/news/2018/11/statement-by-the-organizing-committee-of-the-second-international-summit-on-human-genome-editing
61. Lander, E. S., Baylis, F., Zhang, F., Charpentier, E., Berg, P., et al. (2019). Adopt a moratorium on heritable genome editing. Nature, 567, 165-168. https://doi.org/10.1038/d41586-019-00726-5
62. World Health Organization. (2021, July 12). WHO issues new recommendations on human genome editing for the advancement of public health. https://www.who.int/news/item/12-07-2021-who-issues-new-recommendations-on-human-genome-editing-for-the-advancement-of-public-health
63. The Royal Society. (2023, March). Third International Summit on Human Genome Editing. https://royalsociety.org/science-events-and-lectures/2023/03/2023-human-genome-editing-summit/
64. International Society for Cell & Gene Therapy. (2025). ISCT in the news: joint call for a moratorium on heritable human genome editing. https://www.isctglobal.org/isct-in-the-news-2025
65. MIT Technology Review. (2025, October 31). Here's the latest company planning for gene-edited babies. https://www.technologyreview.com/2025/10/31/1127461/heres-the-latest-company-planning-for-gene-edited-babies/
66. NPR. (2026, September 29). This startup wants to prove gene editing embryos is safe for making healthy babies. https://www.wunc.org/national/2026-09-29/this-startup-wants-to-prove-gene-editing-embryos-is-safe-for-making-healthy-babies