CoChat
Sign Up
Cited by 1,462Cancer Genomics and DiagnosticsAdvanced Proteomics Techniques and ApplicationsGenetic factors in colorectal cancer

Proteogenomic characterization of human colon and rectal cancer

Open access
the NCI CPTAC, Zhang, Bing, Wang, Jing, Wang, Xiaojing, Zhu, Jing, Liu, Qi, Shi, Zhiao, Chambers, Matthew C. +16 more·2014·Nature
DOI 10.1038/nature13438PMC PMC4249766
Open this paper in CoChat — free account

Abstract

Extensive genomic characterization of human cancers presents the problem of inference from genomic abnormalities to cancer phenotypes. To address this problem, we analysed proteomes of colon and rectal tumours characterized previously by The Cancer Genome Atlas (TCGA) and perform integrated proteogenomic analyses. Somatic variants displayed reduced protein abundance compared to germline variants. Messenger RNA transcript abundance did not reliably predict protein abundance differences between tumours. Proteomics identified five proteomic subtypes in the TCGA cohort, two of which overlapped with the TCGA 'microsatellite instability/CpG island methylation phenotype' transcriptomic subtype, but had distinct mutation, methylation and protein expression patterns associated with different clinical outcomes. Although copy number alterations showed strong cis- and trans-effects on mRNA abundance, relatively few of these extend to the protein level. Thus, proteomics data enabled prioritization of candidate driver genes. The chromosome 20q amplicon was associated with the largest global changes at both mRNA and protein levels; proteomics data highlighted potential 20q candidates, including HNF4A (hepatocyte nuclear factor 4, alpha), TOMM34 (translocase of outer mitochondrial membrane 34) and SRC (SRC proto-oncogene, non-receptor tyrosine kinase). Integrated proteogenomic analysis provides functional context to interpret genomic abnormalities and affords a new paradigm for understanding cancer biology.

Integrity signals

Registry

Crossref · 10.1038/nature13438

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 13 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    CNA-protein correlations are much weaker than CNA-mRNA correlations.

    “In contrast, the diagonal pattern was much weaker for CNA-protein correlations (Fig. 3b).”
    Quote in the paperNot a substantial claim
  • Findingown data

    The same trend of higher correlations in focal regions was not observed for focal deletion regions.

    “However, the same trend was not observed for genes in the focal deletion regions.”
    Quote in the paperSubstantial claim
  • Findingown data

    CNA-mRNA correlations were significantly higher than CNA-protein correlations for genes in focal amplification, focal deletion, and non-focal regions.

    “CNA-mRNA correlations were significantly higher than CNA-protein correlations for genes in all three groups (p<1.0 × 10-10, KS test).”
    • Also stated in: § Extended Data · plain
      “CNA-mRNA correlations were significantly higher than CNA-protein correlations for genes in any of the three groups.”
    Quote in the paperSubstantial claim
  • Findingown data

    Genes in focal amplification regions had significantly higher CNA-mRNA and CNA-protein correlations than genes in non-focal regions.

    “genes in the focal amplification regions showed significantly higher CNA-mRNA and CNA-protein correlations than genes in the non-focal regions (p=4.4 × 10-4 and 0.02, respectively, KS test).”
    Quote in the paperSubstantial claim
  • Findingown data

    Somatic variants had lower protein abundance than germline/dbSNP-supported variants.

    “somatic variants exerted a significantly stronger negative impact on protein abundance than did dbSNP-supported variants (p<2.2 × 10-16, Kolmogorov-Smirnov (KS) test, Fig. 1d, Supplementary Methods 5.5).”
    Quote in the paperSubstantial claim
  • May combine two claims. 34
  • The quote only partly states this claim. 2
  • Classified as probably not a substantial claim. 3

  • May combine two claims.

    Findingown data

    In individual samples, steady-state mRNA and protein abundance showed significant positive correlation, with an average correlation of 0.47.

    “All samples showed significant positive mRNA-protein correlation (multiple-test adjusted p value < 0.01, Spearman’s correlation coefficient) and the average correlation between steady state mRNA and protein abundance in individual samples was 0.47 (Fig. 2a)”
    Quote in the paperSubstantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    Across tumors, 89% of genes had positive mRNA-protein correlation, but only 32% had statistically significant correlations.

    “Although 89% of the genes showed a positive mRNA-protein correlation, only 32% had statistically significant correlations (Fig. 2b).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Proteomic subtypes B and C both showed significant association with the TCGA MSI/CIMP subtype, but differ considerably at genomic, epigenomic, and proteomic levels.

    “Proteomic subtypes B and C both showed significant association with the TCGA subtype “MSI/CIMP” (Fig. 4b, Supplementary Table 11); however, they differ considerably at genomic, epigenomic, and proteomic levels (Fig. 4b).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Consensus clustering identified five major proteomic subtypes in the tumor cohort, with 15, 9, 25, 11, and 19 cases in subtypes A through E.

    “Using the Consensus Clustering24 method (Supplementary Methods 9.1–9.2, Extended Data Fig. 9), we identified five major proteomic subtypes in this tumor cohort, with 15, 9, 25, 11, and 19 cases in subtypes A through E, respectively (Fig. 4a–b).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [24] Monti, S., Tamayo, P., Mesirov, J. & Golub, T. R. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data. Mach. Learn. 52, 91–118 (2003) source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    HNF4α abundance was significantly higher in subtype E tumors than in normal colon samples, but not significantly up-regulated in other subtypes.

    “HNF4α abundance was significantly higher in subtype E tumors compared to normal colon samples (multiple-test adjusted p value = 1.09 × 10-6, two-sided Wilcoxon rank-sum test); however, significant up-regulation of HNF4α was not observed for other subtypes (Supplementary Methods 10).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Proteomic subtype C, but not subtype B, showed enriched overlap with the Stem-like subtype and the CCS3 subtype.

    “Proteomic subtype C, but not subtype B, showed enriched overlap with the “Stem-like” subtype described by Sadanandam et al.27 and the “CCS3” subtype described by De Sousa et al.26.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [27] Sadanandam, A. et al. A colorectal cancer classification system that associates cellular phenotype and responses to therapy. Nature Med. 19, 619–625 (2013) source

      Not checked in the free run.

    • cites [26] De Sousa E Melo, F. et al. Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions. Nature Med. 19, 614–618 (2013) source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Almost all hypermutated and MSI-high tumors, as well as tumors with POLE and BRAF mutations, were included in proteomic subtypes B and C.

    “Almost all hypermutated and MSI-high tumors were included in subtypes B and C, as well as tumors with POLE and BRAF mutations.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Subtype E was associated with HNF4A amplification and relatively higher HNF4α protein abundance.

    “Interestingly, subtype E was also associated with HNF4A amplification and relatively higher abundance of HNF4α protein (Fig. 4d).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Subtype B was significantly associated with the TCGA CIMP-H methylation subtype, while subtype C was significantly associated with a non-CIMP subtype (cluster 4).

    “Moreover, whereas subtype B was significantly associated with the TCGA CIMP-H (CIMP-high) methylation subtype, subtype C was significantly associated with a non-CIMP subtype (cluster 4).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Among 79 genes in the 20q region with quantifiable protein measurements, 67 (85%) showed significant CNA-mRNA correlation, but only 40 (51%) showed significant CNA-protein correlation.

    “Among the 79 genes in the 20q region with quantifiable protein measurements, 67 (85%) showed significant CNA-mRNA correlation, but only 40 (51%) showed significant CNA-protein correlation (multiple-test adjusted p value < 0.01, Spearman’s correlation coefficient, Supplementary Table 10).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    Subtype E was significantly associated with both TP53 mutations and 18q loss.

    “In particular, subtype E was significantly associated with both TP53 mutations and 18q loss, genomic features frequently associated with CIN tumors25.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [25] Fearon, E. R. Molecular genetics of colorectal cancer. Annu. Rev. Pathol. 6, 479–507 (2011) source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Subtype B lacked TP53 mutations and chromosome 18q loss.

    “Another unique feature of subtype B was the lack of TP53 mutations and chromosome 18q loss.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    mRNA and protein levels were modestly correlated, but over two thirds of these correlations were not statistically significant in the TCGA tumor set.

    “mRNA and protein levels were modestly correlated, as earlier cell and animal model studies suggest36, but over two thirds of these correlations were not statistically significant in the TCGA tumor set.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [36] Vogel, C. & Marcotte, E. M. Insights into the regulation of protein abundance from proteomic and transcriptomic analyses. Nature Rev. Genet. 13, 227–232 (2012) source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Four network modules were enriched with genes in the up-signature of subtype C, and two modules were enriched with genes in the down-signature.

    “Four network modules were enriched with genes in the up-signature for subtype C, whereas two modules were enriched with genes in the down-signature (multiple-test adjusted p value < 0.01, Fisher’s exact test, Extended Data Fig. 10b).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Four network modules (I, IV, V, VI) were significantly enriched with up-signature proteins and two modules (II and III) were significantly enriched with down-signature proteins (adjusted p < 0.01).

    “Highlighted by red or blue arrows are four Network modules (I, IV, V, VI) significantly enriched with up-signature proteins and two modules (II and III) significantly enriched with down-signature proteins (adjusted p value < 0.01).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Protein bins with spectral counts below 1.4 had clearly lower intraclass correlation coefficients, while the ICC curves began to converge when average spectral count exceeded 1.4.

    “Protein bins with spectral counts less than 1.4 showed clearly lower ICC scores, whereas the ICC score curves started to converge when the average spectral count was greater than 1.4.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Genes with stable mRNA and stable protein had relatively higher mRNA-protein correlation, while genes with unstable mRNA and unstable protein had relatively lower mRNA-protein correlation.

    “Genes with stable mRNA and stable protein showed relatively higher mRNA-protein correlation whereas those with unstable mRNA and unstable protein showed relatively lower mRNA-protein correlation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The spectral count increase after relaxing the false discovery rate occurred for all 7,526 confidently identified proteins, with no protein showing a decrease.

    “Spectral count plot of all 7,526 confidently identified proteins demonstrates the increase in the absolute number of spectra identified for each protein, but no decrease for any of the proteins.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    FPKM-based transcript abundance estimates are independent of gene length, whereas RSEM-based estimates strongly correlate with gene length.

    “FPKM measure is independent of gene length, whereas the RSEM measure strongly correlates with gene length.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    A stronger negative impact of HNF4A shRNA on proliferation was associated with increased copy number, both for the P1-HNF4α-specific shRNA (p=0.04) and for all shRNAs (p=0.01).

    “Moreover, a stronger negative impact was associated with increased copy number, both for the P1- HNF4α specific shRNA (p=0.04, Spearman’s correlation [r]) and for all shRNAs (p=0.01, Spearman’s correlation p-values for individual shRNAs summarized by the Fisher’s combined probability test).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In most scenarios, quantile normalization produced slightly higher intraclass correlation coefficients than global normalization, and both methods clearly outperformed NSAF normalization.

    “In most scenarios, quantile normalization generated slightly higher ICC scores than global normalization, and both methods clearly outperformed the NSAF normalization.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    TRCN0000019193 specifically targets P1 promoter-driven HNF4α isoforms, whereas the other two shRNAs target both P1 and P2 isoforms.

    “TRCN0000019193 specifically targets P1 promoter-driven isoforms, whereas the other two target both types of isoforms.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Among 228,157 exons, 76% were expressed, but only 64% had an average coverage greater than 1.

    “Among all 228,157 exons, 76% were expressed, but only 64% had an average coverage greater than 1.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Approximately 54% of mutation positions were covered by RNA-Seq reads, and only 43% were covered by three or more reads.

    “Approximately 54% of the mutation positions were covered by RNA-Seq reads and only 43% were covered by three or more reads.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Focal amplifications have the strongest cis-effects on both mRNA and protein abundance, suggesting selection for high protein abundance may drive CNA in focal amplification regions.

    “Therefore, focal amplifications have the strongest cis-effects on both mRNA and protein abundance, suggesting that selection for high protein abundance may drive CNA in regions of focal amplification.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Proteomic subtype B is associated with MSI-High and CIMP, but subtype C may have different biological underpinnings.

    “These results clearly established the association between proteomic subtype B and MSI-High and CIMP, but suggest that subtype C might have different biological underpinnings.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Many hot spot-associated protein level alterations occurred without corresponding mRNA alterations, suggesting independent effects at transcriptome and proteome levels.

    “Notably, many hot spot-associated protein level alterations occurred in the absence of corresponding mRNA alterations, suggesting that the same trans-acting hot spot may exert independent effects at both the transcriptome and proteome levels.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    These data strengthen the association of subtype C with poor prognosis and relate it to EMT activation.

    “These data strengthen the association of subtype C with poor prognosis and relate it to EMT activation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Although most CNAs in CRC drive mRNA abundance changes, relatively few translated to consistent changes in protein abundance.

    “Although most CNAs in CRC drive mRNA abundance changes, relatively few translated to consistent changes in protein abundance.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The wound-response gene signature is a powerful predictor of poor clinical outcome in early stage breast cancers.

    “The wound-response gene signature is a powerful predictor of poor clinical outcome in patients with early stage breast cancers29.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [29] Chang, H. Y. et al. Robustness, scalability, and integration of a wound-response gene expression signature in predicting breast cancer survival. Proc. Natl Acad. Sci. USA 102, 3738–3743 (2005) source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Because Stem-like and CCS3 tumors have poor prognosis, proteomic subtype C may also be associated with poor prognosis.

    “Interestingly, tumors with “Stem-like” and “CCS3” classifications both have poor prognosis, which suggests that proteome subtype C also may be associated with poor prognosis.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    Distinguishing subtype B from C through proteomics is important because MSI-High tumors typically have better prognosis.

    “Therefore, the ability to distinguish subtype B from C through proteomics data is important, because MSI-High tumors typically have better prognosis25.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [25] Fearon, E. R. Molecular genetics of colorectal cancer. Annu. Rev. Pathol. 6, 479–507 (2011) source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Proteomics identified CRC subtypes similar to those detectable by transcriptome profiles, but further captured features not detectable in transcript profiles.

    “Proteomics identified CRC subtypes similar to those detectable by transcriptome profiles, but further captured features not detectable in transcript profiles.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    This approach revealed the importance of chromosome 20q amplification and provided new insights into the role of HNF4α in CRC.

    “This approach, which revealed the importance of chromosome 20q amplification and provided new insights into the role of HNF4α in CRC, can be broadly extended to understand roles of CNAs in other cancers.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Interpretationreasoning

    The enrichment of wound-response genes further links subtype C to poor prognosis.

    “This result further links our subtype C to poor prognosis.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given

    Xenograft QC samples were run after every 5 colorectal tumor samples.

    “Xenograft QC samples were run after every 5 colorectal tumor samples.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    No selection criteria other than sample availability were applied for this study.

    “No other selection criteria other than availability were applied for this study.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The rescued peptide-spectrum matches are of high quality and the integrity of the data set is maintained.

    “Rescued PSMs are of high quality with a median PSM FDR score of less than 0.2%, indicating the maintained integrity of the data set.”
    Quote in the paperSubstantial claim
38 of 38

  • The paper relies on this without support.

    The drivers underlying copy number alterations and transcriptional subtypes in colorectal cancer are largely unknown.

    “The drivers underlying copy number alterations (CNAs) and transcriptional subtypes are largely unknown”

  • The paper relies on this without support.

    Proteins link genotypes to phenotypes.

    “Because proteins link genotypes to phenotypes, the Clinical Proteomic Tumor Analysis Consortium (CPTAC) is performing proteomic analyses of TCGA tumor specimens for selected cancer types.”

  • The authors state this as an assumption.

    The area under the CDF increases markedly as k increases until the true number of clusters is reached.

    “As k increases, the area under the CDF is hypothesized to increase markedly until k reaches the ktrue.”
3 of 3

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?

  • CPTAC applied standards and workflows to tumours previously genomically characterized by The Cancer Genome Atlas, pioneered the integration of proteomics with genomics to produce a more unified and comprehensive understanding of cancer biology, and implemented its transition into cancer clinical research studies.

    “CPTAC applied these standards and workflows to tumours previously genomically characterized by The Cancer Genome Atlas. In doing so, CPTAC pioneered the integration of proteomics with genomics (i.e., proteogenomics) to produce a more unified and comprehensive understanding of cancer biology and implemented its transition into cancer clinical research studies96,104,105.”

    From A high-stringency blueprint of the human proteome (2020) · cites this paper as [104]

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Authors
CPTAC, T. N., Zhang, B., Wang, J., Wang, X., Zhu, J., Liu, Q., Shi, Z., Chambers, M. C., Zimmerman, L. J., Shaddox, K. F., Kim, S., Davies, S. R., Wang, S., Wang, P., Kinsinger, C. R., Rivers, R. C., Rodriguez, H., Townsend, R. R., Ellis, M. J. C., Carr, S. A., Tabb, D. L., Coffey, R. J., Slebos, R. J. C., & Liebler, D. C.
Affiliations
  • Vanderbilt University (US)
  • Pacific Northwest National Laboratory (US)
  • Cape Town HVTN Immunology Laboratory / Hutchinson Centre Research Institute of South Africa (ZA)
  • Fred Hutch Cancer Center (US)
  • National Institutes of Health (US)
  • Washington University in St. Louis (US)
  • Harvard University (US)
  • Broad Institute (US)
Corresponding author
Liebler, Daniel C.
ORCID
Venue
Nature
Publisher
Springer Science and Business Media LLC
Published
2014-07-20
Volume
513
Issue
7518
Pages
382-387
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
DOI
10.1038/nature13438
PMC
PMC4249766
OpenAlex
W2131856634
Cited by
1,462
References
81
Citations by year
2026: 76 · 2025: 85 · 2024: 85 · 2023: 117 · 2022: 163 · 2021: 152 · 2020: 132 · 2019: 132 · 2018: 124 · 2017: 139 · 2016: 116 · 2015: 113 · 2014: 26
FWCI
42.10
Type
journal article
Language
en
Open access
Licence
www.springer.com/tdm (tdm)
Locations
  • Nature (publishedVersion)
  • PubMed (publishedVersion)
  • Europe PMC (PubMed Central) (acceptedVersion)
  • PubMed Central (acceptedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • MDC Repository (Max-Delbrueck-Center for Molecular Medicine) (submittedVersion)
  • PMC
Funders
  • National Institutes of Health (5u24ca160035-05, UL1TR000448)
  • National Human Genome Research Institute (U54HG003079)
  • National Cancer Institute (U24CA159988, U24-CA-160035, P50CA095103, U24CA160019, U24CA160034, R01CA046413, P30CA068485)
  • National Institute of General Medical Sciences (GM088822, R01GM088822)
  • National Institute of Diabetes and Digestive and Kidney Diseases (P30DK058404)
  • National Center for Advancing Translational Sciences
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Cancer Research › Cancer Genomics and Diagnostics
  • Physical Sciences › Chemistry › Spectroscopy › Advanced Proteomics Techniques and Applications
  • Health Sciences › Medicine › Pathology and Forensic Medicine › Genetic factors in colorectal cancer
Keywords
colorectal cancer, proteogenomics, protein abundance, proteomic subtypes, copy number alterations, mRNA-protein correlation, driver genes, HNF4α, TOMM34, Src kinase, microsatellite instability, CpG island methylation, clinical outcomes, somatic mutations
MeSH terms
Proto-Oncogene Mas, Mitochondrial Precursor Protein Import Complex Proteins, Chromosomes, Human, Pair 20, Colonic Neoplasms, Humans, Neoplasm Proteins, Rectal Neoplasms, RNA, Messenger, RNA, Neoplasm, Proto-Oncogene Proteins pp60(c-src), Point Mutation, Microsatellite Repeats, CpG Islands, DNA Methylation, Mutation, Missense, Proteome, Genomics, Mitochondrial Membrane Transport Proteins, Proteomics, Hepatocyte Nuclear Factor 4, DNA Copy Number Variations, Transcriptome

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 2026-10-05.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-05.

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