CoChat
Sign Up
Cited by 487Circular RNAs in diseasesRNA Interference and Gene DeliveryRNA and protein synthesis mechanisms

RNA Circularization Diminishes Immunogenicity and Can Extend Translation Duration In Vivo

Open access
Wesselhoeft, R. Alexander, Kowalski, Piotr S., Parker-Hale, Frances C., Huang, Yuxuan, Bisaria, Namita, Anderson, Daniel G.·2019·Molecular Cell
DOI 10.1016/j.molcel.2019.02.015PMC PMC6724735
Open this paper in CoChat — free accountOpen PDF

Abstract

Summary Circular RNAs (circRNAs) are a class of single-stranded RNAs with a contiguous structure that have enhanced stability and a lack of end motifs necessary for interaction with various cellular proteins. Here, we show that unmodified exogenous circRNA is able to bypass cellular RNA sensors and thereby avoid provoking an immune response in RIG-I and toll-like receptor (TLR) competent cells and in mice. The immunogenicity and protein expression stability of circRNA preparations is found to be dependent on purity, with small amounts of contaminating linear RNA leading to robust cellular immune responses. Unmodified circRNA is less immunogenic than unmodified linear mRNA in vitro, in part due to evasion of TLR sensing. Finally, we provide the first demonstration to our knowledge of exogenous circRNA delivery and translation in vivo, and we show that circRNA translation is extended in adipose tissue in comparison to unmodified and uridine-modified linear mRNAs.

AI snapshot

Written from the paper’s full text.

TL;DR

Purified exogenous circRNA bypasses RIG-I and TLR sensing and is less immunogenic than linear mRNA, with extended translation in mouse adipose tissue. Immunogenicity depends on removing linear/triphosphorylated contaminants, while m1ψ and m6A modifications disrupt circRNA production or translation.

Key findings

  • circRNA purity controls immunogenicity: RNase R alone or RNase R plus HPLC was insufficient to prevent cytokine release in A549 cells, but adding phosphatase after HPLC and before RNase R reduced secreted MCP1, IL-6, IFN-α1, TNF-α, and IP-10 to undetectable or untransfected-baseline levels. Late circRNA HPLC fractions elicited no cytokine response above controls, while early circRNA fractions did, and A549 viability at 36 h was significantly greater for late circRNA fractions.
  • Unmodified circRNA was less immunogenic than unmodified linear mRNA in A549 cells despite capping, phosphatase treatment, and HPLC purification of the linear mRNA; unmodified linear mRNA diminished A549 viability, whereas unmodified circRNA did not. TLR3 and TLR8 reporter cells were activated by capped linearized RNA, polyadenylated linearized RNA, nicked circRNA, and early circRNA fractions, but not by late circRNA fractions; adding uridine, but not cytidine, partially reverted TLR8 evasion by circRNA.
  • m1ψ modification disrupted circRNA: complete uridine replacement abolished PIE ribozyme activity, partial replacement dramatically reduced splicing, and splint ligation with T4 RNA ligase I achieved 40% circularization of a 1.5 kb precursor, but m1ψ-circRNA produced no protein expression. m6A modification also impaired function: 10% adenosine replacement reduced splicing of the CVB3 IRES construct but not circ∆IRES, m6A-circRNA had significantly reduced GLuc expression versus unmodified circRNA, and m6A-circ∆IRES produced no detectable GLuc activity.
  • In mice, hEpo was detected in serum after visceral adipose injection; unmodified circRNA decayed more slowly than unmodified or m1ψ-mRNA and was still present 42 h post-injection. Unpurified circRNA splicing reactions and unmodified linear mRNA declined rapidly, and unpurified splicing reactions produced a serum cytokine response not observed for purified circRNA. LNP-delivered circRNA gave lower but comparable serum hEpo to 5moU-mRNA at 6 h, decayed more slowly in adipose tissue, and decayed at almost the same rate as LNP-5moU-mRNA after intravenous liver delivery.

Methods

  • circRNA was synthesized by permuted intron-exon (PIE) splicing with CVB3 IRES and GLuc or hEpo coding regions, then purified by RNase R, HPLC, and/or phosphatase; m1ψ-circRNA was made by splint ligation with T4 RNA ligase I, and circRNA was linearized by heat/Mg2+ or RNase H nicking.
  • Cell work used 293, A549, HeLa, RAW264.7, and HEK-Blue TLR3/TLR7/TLR8/null reporter cells; readouts included cytokine immunoassays, GLuc luminescence, hEpo ELISA, qPCR for RIG-I/IFN-β1/IL-6/RANTES/TNF-α, cell viability, and SEAP reporter activity.
  • Mice were 30–35 g C57Bl/6 females injected into visceral fat/mammary fat pad and peritoneum with 350 ng RNA complexed with MessengerMax or 1.5 pmol LNP-RNA, or intravenously with 0.1 mg/kg LNP-RNA; LNPs used cKK-E12, DOPE, cholesterol, and C14-PEG 2000 at a 35:16:46.5:2.5 molar ratio. Statistics used two-tailed unpaired Welch’s t-test with p < 0.05.

Limitations

  • Stated: the CVB3 IRES may not be ideal for mouse adipose or liver tissue, IRES sequences compete with capped endogenous transcripts for initiation factors, and comprehensive characterization of other IRES sequences across tissues is needed. The paper also states that the mechanism of TLR evasion remains unclear, no degradation product has been defined as a TLR3 ligand, and biochemical experiments on sensor-circRNA interactions are warranted.
  • Stated: circRNA stability was tissue-specific, with no expression-stability advantage over 5moU-mRNA after intravenous LNP delivery to liver; factors such as RNA turnover, endonuclease activity, sequence-specific translation inhibition/degradation, and RNA-stabilizing proteins were not resolved.
  • Evident from the text: the statistical analysis section states that data presented are representative of one independent experiment, so reproducibility across independent replicates is not demonstrated in the provided text.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.molcel.2019.02.015

Cites

38 of 38 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 7 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    The protein expression stability of circRNA preparations depends on purity.

    “The immunogenicity and protein expression stability of circRNA preparations is found to be dependent on purity, with small amounts of contaminating linear RNA leading to robust cellular immune responses.”
    • Also stated in: § Purification of exogenous circRNA ablates immunogenicity · plain
      “The stability of protein expression from circRNA is also dependent on circRNA purity and the sensitivity of transfected cell types to contaminating RNA species.”
    Quote in the paperSubstantial claim
  • Findingown data

    HPLC purification alone is insufficient to prevent cytokine release.

    “The addition of HPLC purification was furthermore insufficient to prevent cytokine release”
    Quote in the paperSubstantial claim
  • Findingown data

    Purified circRNA containing an EMCV IRES and EGFP coding region failed to induce substantial pro-inflammatory transcripts in A549 cells.

    “we tested the induction of pro-inflammatory transcripts in response to transfection of A549 cells with purified circRNA containing an EMCV IRES and EGFP coding region, and again failed to observe substantial induction”
    Quote in the paperSubstantial claim
  • Findingown data

    Purified circRNA failed to induce pro-inflammatory transcripts in RAW264.7 murine macrophages.

    “Purified circRNA similarly failed to induce pro-inflammatory transcripts in RAW264.7 murine macrophages”
    Quote in the paperSubstantial claim
  • Findingown data

    Small amounts of contaminating linear RNA lead to robust cellular immune responses.

    “with small amounts of contaminating linear RNA leading to robust cellular immune responses.”
    • Also stated in: § Non-circular components of the splicing reaction contribute to immunogenicity · hedged
      “indicating that even relatively small quantities of linear RNA contaminants are able to induce a substantial cellular immune response in A549 cells.”
    • Also stated in: § Discussion · hedged
      “even small quantities of contaminating linear RNA, some of which may harbor triphosphates and may be present after HPLC purification, are sufficient to provoke robust cellular immune responses”
    Quote in the paperSubstantial claim
  • May combine two claims. 23
  • Classified as probably not a substantial claim. 8

  • May combine two claims.

    Findingown data

    Increased circRNA purity improved GLuc expression stability in transfected A549 cells, with completely purified circRNA demonstrating a stability phenotype similar to transfected 293 cells.

    “increased circRNA purity improved GLuc expression stability in transfected A549 cells, with completely purified circRNA demonstrating a stability phenotype similar to that of transfected 293 cells”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Exogenous circRNA delivery and translation in vivo was demonstrated.

    “we provide the first demonstration to our knowledge of exogenous circRNA delivery and translation in vivo”
    • Also stated in: § Exogenous circRNA is translatable in vivo · plain
      “In mice, hEpo was detected in serum after injection of hEpo circRNA or linear mRNA into visceral adipose”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The hEpo detected in serum after injection reflects protein translated from the injected RNA.

  • May combine two claims.

    Findingown data

    In HeLa cells, transfection of RNase R-digested splicing reactions, but not purified circRNA, induced RIG-I and IFN-β1.

    “In HeLa cells, transfection of RNase R-digested splicing reactions, but not purified circRNA, induced RIG-I and IFN-β1”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    A trend of increased circRNA purity improved A549 cell viability 3 days post-transfection while 293 cell viabilities remained largely unaffected.

    “a trend of increased circRNA purity improved A549 cell viability 3 days post-transfection while 293 cell viabilities remained largely unaffected”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Compared to unpurified splicing reaction, combined RNase R and HPLC caused a significant reduction in IL-6 and a significant increase in IFN-α1.

    “although we did note a significant reduction in IL-6 and a significant increase in IFN-α1 compared to the unpurified splicing reaction”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The late circRNA fraction did not provoke a TLR-mediated response in any cell line, similarly to m1ψ-mRNA.

    “the late circRNA fraction did not provoke a TLR-mediated response in any cell line, similarly to m1ψ-mRNA”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    HPLC purification of circRNA presents unique difficulties because nicked circRNA and intact circRNA are equal in molecular weight and their respective peaks partly overlap.

    “HPLC purification of circRNA presents unique difficulties, as nicked circRNA and intact circRNA are equal in molecular weight and their respective peaks partly overlap.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    A549 cell viability was diminished upon transfection of unmodified linear mRNA, but not unmodified circRNA or either m1ψ-RNAs.

    “A549 cell viability was diminished upon transfection of unmodified linear mRNA, but not unmodified circRNA or either m1ψ-RNAs”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Encapsulation in LNPs did not alter RIG-I/IFN-β1 induction or TLR activation in vitro, with unmodified circRNA failing to activate immune sensors in a manner similar to 5moU-mRNA.

    “Encapsulation in LNPs did not alter RIG-I/IFN-β1 induction or TLR activation in vitro, with unmodified circRNA failing to activate immune sensors in a manner similar to 5moU-mRNA”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    RNase R treatment is not sufficient to obtain pure circRNA and enriches multiple resistant RNA species, including circRNA and linear RNAs with structured 3' ends.

    “We find that treatment with RNase R is not sufficient to obtain pure circRNA and enriches multiple resistant RNA species, which include circRNA and linear RNAs with structured 3’ ends.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Modification of circRNA precursor molecules with m1ψ and m6A interfered with splicing in the PIE constructs and translation in the enzymatically circularized RNAs.

    “Modification of circRNA precursor molecules with m1ψ and m6A interfered with splicing in the PIE constructs and translation in the enzymatically circularized RNAs”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both heat and RNase H linearization yielded a majority of full-length linear RNA with small amounts of intact circRNA.

    “Both methods yielded a majority of full-length linear RNA with small amounts of intact circRNA”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Heat treatment resulted in a greater proportion of lower molecular weight linear RNA degradation products than RNase H digestion.

    “although heat treatment resulted in a greater proportion of lower molecular weight linear RNA degradation products”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    No increase in serum cytokines, or local RIG-I, TNF-α, or IL-6 transcript induction was observed after injection of LNP-5moU-mRNA or LNP-circRNA.

    “We did not observe an increase in serum cytokines, or local RIG-I, TNF-α, or IL-6 transcript induction after injection of LNP-5moU-mRNA or LNP-circRNA”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    No significant differences in 293 cytokine release at 24 hours post-transfection and cell viability at 3 days post-transfection were detected.

    “we did not detect significant differences in 293 cytokine release at 24 hours post-transfection and cell viability at 3 days post-transfection”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Addition of uridine, but not cytidine, to the media of TLR8 reporter cells transfected with circRNA partially reverted the effect and resulted in SEAP secretion.

    “the addition of uridine, but not cytidine, to the media of TLR8 reporter cells transfected with circRNA partially reverted this effect and resulted in SEAP secretion”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    10% replacement of adenosine with m6A reduced splicing of the longer circRNA construct containing the CVB3 IRES, but did not impact splicing of circ∆IRES.

    “10% replacement of adenosine with m6A reduced splicing of the longer circRNA construct containing the CVB3 IRES, but did not impact splicing of circ∆IRES”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Using the purification protocol described, circRNA does not elicit substantial innate immune responses from TLR and RIG-I competent cells, in contrast to other components of the splicing reaction, or from mouse adipose tissue, despite the absence of circRNA-associated host splicing factors.

    “Using the purification protocol described here, we find that circRNA does not elicit substantial innate immune responses from TLR and RIG-I competent cells, in contrast to other components of the splicing reaction, or from mouse adipose tissue, despite the absence of circRNA-associated host splicing factors.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both m1ψ-circRNA and m1ψ-mRNA did not significantly alter cytokine release profiles.

    “both m1ψ-circRNA and m1ψ-mRNA did not significantly alter cytokine release profiles”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    hEpo linear mRNA and circRNA displayed similar relative protein expression patterns and cell viabilities in comparison to GLuc linear mRNA and circRNA upon equal weight transfection of 293 and A549 cells.

    “hEpo linear mRNA and circRNA displayed similar relative protein expression patterns and cell viabilities in comparison to GLuc linear mRNA and circRNA upon equal weight transfection of 293 and A549 cells”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In vitro transcribed linearized circRNA elicited a TLR3-mediated response, whereas linearized circRNA produced by degrading purified circRNA did not.

    “We furthermore observed differences in TLR3 activation by linearized circRNA, with in vitro transcribed linearized circRNA eliciting a TLR3-mediated response while linearized circRNA produced by degrading purified circRNA did not (Figures 4B,F, Figure S4C).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Unmodified exogenous circRNA bypasses cellular RNA sensors and avoids provoking an immune response in RIG-I and TLR competent cells and in mice.

    “Here, we show that unmodified exogenous circRNA is able to bypass cellular RNA sensors and thereby avoid provoking an immune response in RIG-I and toll-like receptor (TLR) competent cells and in mice.”
    • Also stated in: § Discussion · plain
      “we have found that exogenous circRNA does not activate several known cellular RNA sensors including TLRs and RIG-I.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Previous studies used circRNA purified by RNase R.

    “Previous studies have used circRNA purified by RNase R (Chen et al. 2017).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Chen et al. 2017 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    Improved cell viability and protein production stability are indicators of an antiviral response to non-circular contaminants.

    “both of which are indicators of an antiviral response to non-circular contaminants”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Loo & Gale 2011

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Incorporation of ψ enhances base-stacking interactions, which may lead to structural alterations.

    “Incorporation of ψ has been shown to enhance base-stacking interactions, which may lead to structural alterations (Davis 1995)”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Davis 1995 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given

    No degradation product has yet been defined as a ligand for TLR3.

    “However, no degradation product has yet to be defined as a ligand for TLR3”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Mice were randomly assigned to treatment or control groups.

    “30–35g C57Bl/6 female mice (Charles River) randomly assigned to treatment or control groups were injected into visceral fat through the lower right mammary fat pad and peritoneum with 350ng of RNA complexed with MessengerMax or 1.5 picomoles of LNP-RNA in a total volume of 50µL, or intravenously by tail vein injection with 0.1mg/kg LNP-RNA.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    For all studies, data presented is representative of one independent experiment.

    “For all studies, data presented is representative of one independent experiment.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The CVB3 IRES was originally selected for use in circRNA based on its ability to drive translation in human cell lines.

    “The CVB3 IRES was originally selected for use in circRNA based on its ability to drive translation in human cell lines (Wesselhoeft et al. 2018).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Wesselhoeft et al. 2018 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    293 and HeLa cells tested negative for mycoplasma.

    “293 and HeLa cells tested negative for mycoplasma.”
    Quote in the paperNot a substantial claim
30 of 30

  • The authors state this as an assumption.

    The contribution of TLRs to circRNA immunogenicity and the effects of nucleoside modifications including m1ψ and m6A on exogenous circRNA translation, stability, and immunogenicity had not been reported.

    “To our knowledge, the contribution of TLRs to circRNA immunogenicity, and the effects of nucleoside modifications including m1ψ and m6A on exogenous circRNA translation, stability, and immunogenicity, have yet to be reported.”

  • The paper relies on this without support.

    The hEpo detected in serum after injection reflects protein translated from the injected RNA.

    “In mice, hEpo was detected in serum after injection of hEpo circRNA or linear mRNA into visceral adipose”
2 of 2

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Unmodified circRNA has been shown to induce TLR/RIG-I-mediated innate immune response compared to unmodified linear mRNA.

    “unmodified circRNA has been shown to induce TLR/RIG-I-mediated innate immune response compared to unmodified linear mRNA.225,226”

    From Advances in COVID-19 mRNA vaccine development (2022) · cites this paper as [225]

  • Circular RNA without a stop codon reduces the frequency of ribosome detachment from the RNA, thereby enabling continuous translation and high protein expression.

    “Moreover, circRNA without a stop codon reduces the frequency of ribosome detachment from the RNA, thereby enabling continuous translation and high protein expression.”

    From Lipid nanoparticles for mRNA delivery (2021) · cites this paper as [246]

  • Synthetic circular RNA can be made by covalently linking the 3′ and 5′ ends of a linear precursor using enzymatic or chemical methods.

    “Synthetic circRNA can be made by covalently linking the 3′ and 5′ ends of a linear precursor using enzymatic or chemical methods.”

    From Lipid nanoparticles for mRNA delivery (2021) · cites this paper as [246]

  • Circular RNA is a single-stranded RNA with a closed-loop structure.

    “Circular RNA (circRNA) is a single-stranded RNA with a closed-loop structure.”

    From Lipid nanoparticles for mRNA delivery (2021) · cites this paper as [246]

  • Chemical modification of specific nucleotides and chromatographic purification can minimize immunogenicity of RNA and increase translation.

    “Similar to linear mRNA, the chemical modification of specific nucleotides and chromatographic purification can minimize immunogenicity of the RNA and increase translation.”

    From Lipid nanoparticles for mRNA delivery (2021) · cites this paper as [246]

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Wesselhoeft, R. A., Kowalski, P. S., Parker-Hale, F. C., Huang, Y., Bisaria, N., & Anderson, D. G.
Affiliations
  • Koch Institute for Integrative Cancer Research At MIT (US)
  • Massachusetts Institute of Technology (US)
  • Whitehead Institute for Biomedical Research (US)
  • Harvard University (US)
Corresponding author
Anderson, Daniel G.
ORCID
Venue
Molecular Cell
Publisher
Elsevier BV
Published
2019-05
Volume
74
Issue
3
Pages
508-520.e4
ISSN
1097-2765 (ISSN-L)
DOI
10.1016/j.molcel.2019.02.015
PMC
PMC6724735
OpenAlex
W2923639686
Semantic Scholar
76856cd95b9e255a7e0eb4712a223ed8dab381ca
Cited by
482
References
38
Citations by year
2026: 91 · 2025: 99 · 2024: 81 · 2023: 61 · 2022: 59 · 2021: 47 · 2020: 31 · 2019: 11
FWCI
12.66
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecular Cell (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc-nd)
Funders
  • NIH (5R01HL125428)
  • National Cancer Institute (P30-CA14051)
  • MIT School of Science
  • Juvenile Diabetes Research Foundation (3-PDF-2017-383-A-N)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Circular RNAs in diseases
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › RNA Interference and Gene Delivery
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › RNA and protein synthesis mechanisms
Keywords
circular RNA, RNA circularization, immunogenicity, innate immune evasion, RIG-I, Toll-like receptors, RNA sensors, circRNA translation, adipose tissue, RNA stability
MeSH terms
DEAD Box Protein 58, RNA, Circular, Animals, Gene Expression Regulation, Immunity, Innate, RNA, RNA, Messenger, Protein Biosynthesis, Uridine, Vaccines, Synthetic, MicroRNAs, Toll-Like Receptors, Mice, Gene Regulatory Networks

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review