CoChat
Sign Up
Cited by 7,107Bioinformatics and Genomic NetworksGenomics and Phylogenetic StudiesGene expression and cancer classification

The DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.

Gene ontology analysis for RNA-seq: accounting for selection bias

Open access
Young, Matthew D, Wakefield, Matthew J, Smyth, Gordon K, Oshlack, Alicia·2010·Genome Biology
DOI 10.1186/gb-2010-11-2-r14PMC PMC2872874
Open this paper in CoChat — free accountOpen PDF

Abstract

We present GOseq, an application for performing Gene Ontology (GO) analysis on RNA-seq data. GO analysis is widely used to reduce complexity and highlight biological processes in genome-wide expression studies, but standard methods give biased results on RNA-seq data due to over-detection of differential expression for long and highly expressed transcripts. Application of GOseq to a prostate cancer data set shows that GOseq dramatically changes the results, highlighting categories more consistent with the known biology.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1186/gb-2010-11-2-r14

Cites

30 of 30 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Length distribution of genes in Gene Ontology categories. (a) The distribution of average gene lengths in GO categories on a log10 scale. The GO category gene length is given by the median length of the genes within the category. (b) P-values for the two-sided Mann-Whitney U test comparing the median length of genes in a GO category with the overall distribution of genes for 7,873 GO categories. The excess of low P-values shows that there are many GO categories that contain a set of significantly long or short genes.
Figure 1 Length distribution of genes in Gene Ontology categories. (a) The distribution of average gene lengths in GO categories on a log10 scale. The GO category gene length is given by the median length of the genes within the category. (b) P-values for the two-sided Mann-Whitney U test comparing the median length of genes in a GO category with the overall distribution of genes for 7,873 GO categories. The excess of low P-values shows that there are many GO categories that contain a set of significantly long or short genes.CC BY · PubMed Central
Figure 2 Differential expression as a function of gene length and read count. (a) The proportion of DE genes is plotted as a function of the transcript length. Each point represents the percentage of genes called DE in a bin of 300 genes plotted against the median gene length for the bin. The green line is the probability weighting function obtained by fitting a monotonic cubic spline with 6 knots to the binary data. A clear trend towards more detected differential expression at longer gene lengths can be seen. (b) The same, except instead of transcript length, the total number of reads for each gene was used. Again, a trend towards more DE for genes with more reads can be seen. Note the greater range of probabilities compared to (a).
Figure 2 Differential expression as a function of gene length and read count. (a) The proportion of DE genes is plotted as a function of the transcript length. Each point represents the percentage of genes called DE in a bin of 300 genes plotted against the median gene length for the bin. The green line is the probability weighting function obtained by fitting a monotonic cubic spline with 6 knots to the binary data. A clear trend towards more detected differential expression at longer gene lengths can be seen. (b) The same, except instead of transcript length, the total number of reads for each gene was used. Again, a trend towards more DE for genes with more reads can be seen. Note the greater range of probabilities compared to (a).CC BY · PubMed Central
Figure 3 Change in Gene Ontology category rank between the standard and GOseq methodologies. (a) Change in rank of GO categories going from the hypergeometric method to GOseq correcting for length bias plotted against the log of the average gene length of the category. (b) Change in rank of GO categories going from the hypergeometric method to GOseq correcting for total read count plotted against the log of the average number of counts of each gene in the category. A trend for the standard method to underestimate significance for GO categories containing short (or highly expressed) genes and overestimate significance for GO categories containing long (or underexpressed) genes can be clearly seen.
Figure 3 Change in Gene Ontology category rank between the standard and GOseq methodologies. (a) Change in rank of GO categories going from the hypergeometric method to GOseq correcting for length bias plotted against the log of the average gene length of the category. (b) Change in rank of GO categories going from the hypergeometric method to GOseq correcting for total read count plotted against the log of the average number of counts of each gene in the category. A trend for the standard method to underestimate significance for GO categories containing short (or highly expressed) genes and overestimate significance for GO categories containing long (or underexpressed) genes can be clearly seen.CC BY · PubMed Central
Figure 4 Comparison of GOseq and the standard hypergeometric methods. (a) The P-values generated with GOseq using the Wallenius approximation and the standard hypergeometric method are plotted against the P-values calculated with GOseq using random sampling (200k repeats). The Wallenius method (green crosses) shows good agreement with the high resolution (200,000 repeats) random sampling. A large discrepancy in P-values is seen between GOseq and the hypergeometric method. (a) The number of discrepancies between lists is shown for a given list size. The black line compares GOseq using high resolution sampling with the hypergeometric method. The red line compares GOseq using high resolution sampling with the Wallenius approximation. Again, GOseq using the Wallenius method shows little difference from GOseq using the random sampling method (with 200k repeats) while the hypergeometric method shows a large number (approximately 20%) of discrepancies.
Figure 4 Comparison of GOseq and the standard hypergeometric methods. (a) The P-values generated with GOseq using the Wallenius approximation and the standard hypergeometric method are plotted against the P-values calculated with GOseq using random sampling (200k repeats). The Wallenius method (green crosses) shows good agreement with the high resolution (200,000 repeats) random sampling. A large discrepancy in P-values is seen between GOseq and the hypergeometric method. (a) The number of discrepancies between lists is shown for a given list size. The black line compares GOseq using high resolution sampling with the hypergeometric method. The red line compares GOseq using high resolution sampling with the Wallenius approximation. Again, GOseq using the Wallenius method shows little difference from GOseq using the random sampling method (with 200k repeats) while the hypergeometric method shows a large number (approximately 20%) of discrepancies.CC BY · PubMed Central
Figure 5 A comparison of Gene Ontology analysis using RNA-seq and microarrays on the same samples. The fraction of GO categories identified by RNA-seq data that overlap with the microarray GO analysis are shown as a function of the number of categories selected. RNA-seq data have been analyzed using GOseq and hypergeometric methods. The GOseq categories have a consistently higher overlap with the microarray GO categories than the standard method.
Figure 5 A comparison of Gene Ontology analysis using RNA-seq and microarrays on the same samples. The fraction of GO categories identified by RNA-seq data that overlap with the microarray GO analysis are shown as a function of the number of categories selected. RNA-seq data have been analyzed using GOseq and hypergeometric methods. The GOseq categories have a consistently higher overlap with the microarray GO categories than the standard method.CC BY · PubMed Central
Figure 6 Change in most significant categories when correcting for total read count bias or length bias versus the standard method. A plot of the number of discrepancies in the most significant GO categories generated using different methods. This plot compares the length bias correcting version of GOseq to the standard hypergeometric method (green line) and the total read count bias correcting version of GO-seq to the standard hypergeometric method (black line).
Figure 6 Change in most significant categories when correcting for total read count bias or length bias versus the standard method. A plot of the number of discrepancies in the most significant GO categories generated using different methods. This plot compares the length bias correcting version of GOseq to the standard hypergeometric method (green line) and the total read count bias correcting version of GO-seq to the standard hypergeometric method (black line).CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 30 references

Research works printed but not linked: 1

  1. Barnard GA: Contribution to the discussion of Professor Bartlet's paper. J Roy Stat Soc Series B (Methodological). 1963, 25: 294-

Cited by

Sorted by citations

Items 1,001–1,009 of 1,009

Related

See all

Metadata

Authors
Young, M. D., Wakefield, M. J., Smyth, G. K., & Oshlack, A.
Affiliations
Walter and Eliza Hall Institute of Medical Research (AU)
Corresponding author
Oshlack, Alicia
ORCID
Venue
Genome Biology
Publisher
Springer Science and Business Media LLC
Published
2010-02-04
Volume
11
Issue
2
Pages
R14-R14
ISSN
  • 1465-6906 (ISSN-L)
  • 1474-760X
DOI
10.1186/gb-2010-11-2-r14
PMC
PMC2872874
OpenAlex
W2107060225
Cited by
7,107
References
30
Citations by year
2026: 284 · 2025: 495 · 2024: 690 · 2023: 713 · 2022: 918 · 2021: 1008 · 2020: 849 · 2019: 657 · 2018: 476 · 2017: 385 · 2016: 229 · 2015: 160 · 2014: 95 · 2013: 58 · 2012: 44
FWCI
18.77
Type
journal article
Language
en
Open access
Licence
Locations
  • Genome biology (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Minerva Access (University of Melbourne) (submittedVersion, cc-by)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
Keywords
gene ontology enrichment, RNA-seq, selection bias, differential gene expression, prostate cancer, gene expression analysis
MeSH terms
Androgens, Humans, Male, Prostatic Neoplasms, Gene Expression Regulation, Neoplastic, Bias, Sequence Analysis, RNA, Cell Line, Tumor, Genome-Wide Association Study

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.

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