Differential Expression Analysis for Sequence Count Data
Nature Precedings, 2010 · paper · Biomedicine
Proposes a method for detecting differential signal in sequencing count data, using a negative binomial model with variance and mean linked by local regression. It is implemented as the DESeq R/Bioconductor package.
Checked: PDF from europepmc.org, fetched 4 Oct 2026. Page numbers are those of that PDF.
7 findings
Substance Page 3, Equation (12): the "pooled mean estimate" is printed as q̂_i0 = Σ_{j:ρ(j)∈{A,B}} k_ij / s_j, a plain sum with no division by the number of samples.
- p. 3 Equation (12) and the sentence before it
pute the pooled mean estimate from the counts of both conditions
- p. 3 Equation (6)
To estimate qiρ, we use the average of the counts from the samples j corresponding to condition ρ
Judge's note: The page 3 image shows Equation (12) as a sum over the samples of both conditions with no 1/m factor, while Equation (6) defines the per-condition mean with 1/mρ. Equations (13) and (14) on page 4 multiply q̂_i0 by s_j and sum over the samples of one condition, which is only right if q̂_i0 is a mean. As printed, the summed mean would be too large by a factor of mA+mB. I found no statement in the text that redefines q̂_i0 as a total. The example in the arithmetic uses size factors of 1, two samples per condition and count 1 per sample: Equation (12) gives 4, and Equation (13) then gives 8 for condition A where the expected total is 2.
Second check: another model agreed · checked by hand
Report a problem with this finding
Substance The DESeq hit count for the full fly analysis is "864 genes (of 17,605)" on page 4 and "DESeq reported 864 hits" on page 6, but "only 90 of the 862 hits (11%)" on page 6.
- p. 4 Testing
for 864 genes (of 17,605)
- p. 6 Comparison with edgeR
DESeq reported 864 hits
- p. 6 Working without replicates
only 90 of the 862 hits (11%) were recovered
- p. 6 Working without replicates
only 90 of the 862 hits (11%)
Only the authors can say what is right here.
Judge's note: Checked the page 4 and page 6 images; all three numbers are printed as quoted. The no-replicates paragraph refers to the hits "from the more reliable analysis with all available samples", which is the same A against B analysis at 10% FDR. I searched the text for a different analysis set, filter or software setting that would give 862 and found none. The "two new hits" are hits outside the full list and do not explain the difference. Only the authors can say which number is right. The 11% is treated in C013.
Second check: another model agreed · checked by hand
Report a problem with this finding
Reference or label Figure 5 caption reads "Sample clustering for the neural cell data of Kasowski et al. [18]"; reference 18 is Engström et al., and Kasowski et al. is reference 20 (the ChIP-Seq data).
- p. 8 Figure 5 caption
Sample clustering for the neural cell data of Kasowski et al. [18]
- p. 4 Applications, Data sets
Engström et al. [18] performed Tag-Seq [19] for tissue cultures of neural cells
- p. 11 Reference 18
Engström P, Tommei D, Stricker S, Smith A, Pollard S, Bertone P
- p. 8 Figure 5 caption
neural cell data of Kasowski et al. [18]
Judge's note: Page 8 image: the caption prints "Kasowski et al. [18]" and the figure shows GNS/NS samples (GliNS1, G144, CB660, CB541, G166, G179). Page 11 image: reference 18 is Engström P, Tommei D, Stricker S, Smith A, Pollard S, Bertone P; reference 20 is Kasowski et al., Science 2010. The Data sets text (page 4) says "Engström et al. [18] performed Tag-Seq" for the neural stem cells and "Kasowski et al. [20]" for the HapMap ChIP-Seq. The reference number is right and the author name is wrong.
Second check: another model agreed
Report a problem with this finding
Reference or label Figure 3 caption: "Supplementary Figure S8 in Additional file 2" and page 10: "Supplementary Figure S3 in Additional file 2"; the Additional material list puts all Supplementary Figures in Additional file 1.
- p. 6 Figure 3 caption
Supplementary Figure S8 in Additional file 2
- p. 11 Additional material
Additional file 1: Supplement. Contains all Supplementary Notes and Supplementary Figures.
- p. 10 Conclusions, paragraph beginning "Fourth"
Plots such as Supplementary Figure S3 in Additional file 2 are helpful
- p. 10 Fourth point
Supplementary Figure S3 in Additional file 2
Judge's note: Checked pages 5, 6, 10 and 11. The list on page 11 reads "Additional file 1: Supplement. Contains all Supplementary Notes and Supplementary Figures" and describes Additional file 2 as the Sweave source-code document. The Figure 1 caption places Supplementary Figure S9 in Additional file 1, and every Supplementary Note pointer uses file 1. The two pointers to file 2 are therefore inconsistent with the paper's own description. The supplement files were not provided, so I could not open S3 and S8 to confirm where they sit.
Second check: another model agreed
Report a problem with this finding
Reference or label Two pointers place supplementary figures in Additional file 2 ("Supplementary Figure S8 in Additional file 2", Figure 3 caption; "Supplementary Figure S3 in Additional file 2", page 10), but the Additional material list says Additional file 1 "Contains all Supplementary Notes and Supplementary Figures" and file 2 is the source-code document.
- p. 10 Sharing of information between genes, 'Fourth' paragraph
Plots such as Supplementary Figure S3 in Additional file 2 are helpful
- p. 11 Additional material
Additional file 1: Supplement. Contains all Supplementary Notes and Supplementary Figures.
- p. 6 Figure 3 caption, last sentence
The corresponding volcano plot is shown in Supplementary Figure S8 in Additional file 2.
- p. 5 Figure 1 caption, last sentence
see Supplementary Figure S9 in Additional file 1
Judge's note: Page 6 image (Figure 3 caption) and page 10 image ("Fourth" paragraph) both print Additional file 2. Page 11 image: Additional file 1 is the Supplement with all Supplementary Notes and Figures; Additional file 2 is "Supplement II. PDF file presenting the source code". The Figure 1 caption cites Supplementary Figure S9 in Additional file 1, and all Supplementary Notes are cited in Additional file 1. The supplement files were not provided, so I could not open them, but the article contradicts itself and the evident fix is Additional file 1.
Second check: another model agreed
Report a problem with this finding
Wording Page 4: "As each tissue culture was derived from a different subject and so has a different genotype" is contradicted by the Figure 5 caption: "Two GNS samples were derived from the same patient (marked with '(*)')".
- p. 4 Data sets, Tag-Seq
each tissue culture was derived from a different subject
- p. 8 Figure 5 caption
Two GNS samples were derived from the same patient
Judge's note: Checked the page 4 and page 8 images. Figure 5 shows the six neural samples (four GNS, two NS), which is the Tag-Seq data set described on page 4, and two GNS rows carry the '(*)' mark. The data description therefore overstates: not every culture came from a different subject. The caption is the more specific statement, so the page 4 wording is the one to fix. Separately, the Figure 5 title attributes the neural cell data to "Kasowski et al. [18]", but reference 18 is Engström et al. and Kasowski is reference 20; that is outside this cluster.
Second check: another model agreed
Report a problem with this finding
Wording Figure 5 caption says the heat map runs "from dark blue for zero distance to orange for large distance"; the colour key and the heat map run from dark blue to yellow, with no orange.
- p. 8 Figure 5 caption
dark blue for zero distance to orange for large distance
- p. 8 Figure 5, Color Key
Color Key
Judge's note: Page 8 image: the Color Key is a gradient from dark blue (0) through olive to bright yellow (above 200), and the large-distance cells are yellow. The caption's "orange" does not match the printed figure. Colour naming is somewhat subjective, so this is a minor caption-versus-figure mismatch.
Second check: another model agreed
Report a problem with this finding
15 minor (typos and the like; most checked by the judge only)
Reference or label "(as recommended by Robinson et al. [8] and others)" cites reference 8, which has two authors (Robinson MD, Smyth GK); the three-author edgeR paper (Robinson, McCarthy, Smyth) is reference 10.
- p. 9 Sharing of information between genes
mended by Robinson et al. [8] and others)
- p. 11 Reference 8
Robinson MD, Smyth GK: Moderated statistical tests for assessing differences in tag abundance
- p. 9 Sharing of information between genes, first sentence
mended by Robinson et al. [8] and others) may result in
- p. 9 Choice of distribution, last paragraph
published methods using this distribution is likely edgeR [8]
Judge's note: Page 9 image shows "mended by Robinson et al. [8]" and "edgeR [8]". Page 11 image: reference 8 is Robinson MD, Smyth GK, Bioinformatics 2007; reference 10 is Robinson MD, McCarthy DJ, Smyth GK, the edgeR package paper. Elsewhere the paper writes "Robinson and Smyth [11]" for a two-author work. Either the name form or the number is off, but reference 8 is one of the methods papers behind edgeR, so citing it is defensible. Only the authors can say which was meant.
Second check: another model agreed
Report a problem with this finding
Wording Page 4 gives "the shot noise ŝ_j q̂_iρ" and the Figure 1 caption gives "that is, ŝ_j q̂_i,A" for lines on a common-scale variance plot, where the Poisson variance is q̂/ŝ_j as in Equation (8).
- p. 4 Variance estimation, first paragraph
regression fit wρ(q) and the shot noise
- p. 5 Figure 1 caption
variance implied by the Poisson distribution for each of the two samples, that is,
- p. 3 Equation (8)
and define
Only the authors can say what is right here.
Judge's note: Checked the page 3, 4 and 5 images. Figure 1a plots common-scale sample variances (Equation (7)) against common-scale means. The Poisson variance of k_ij is s_j·q; divided by s_j² for the common scale it is q/s_j, which is the form in Equation (8) (z = q̂/m · Σ 1/ŝ_j). The text and the caption instead give ŝ_j·q̂, the raw-scale variance. With size factors near 1 the plotted lines would look almost the same, so this is a notational inconsistency and not a visible error in the figure. The authors would need to say which expression was plotted.
Second check: another model agreed
Report a problem with this finding
Wording Page 6: "only 90 of the 862 hits (11%) were recovered (with two new hits)"; 90 of 862 is 10.4%, which rounds to 10%.
- p. 6 Working without replicates
only 90 of the 862 hits (11%) were recovered (with two new hits)
- p. 6 Working without replicates, last paragraph
only 90 of the 862 hits (11%) were recovered
- p. 6 Working without replicates
only 90 of the 862 hits (11%)
Only the authors can say what is right here.
Judge's note: Checked the page 6 crop: the text is printed as quoted. 90/862 is 10.44% and 90/864 (the count given elsewhere) is 10.42%; neither rounds to 11%. Adding the two new hits to the numerator (92/862 = 10.67%) would give 11%, but the sentence says 90 were recovered and the new hits are not among the 862. The difference is one percentage point and does not change the conclusion, so it is rated minor. The 862 versus 864 mismatch is C010.
Second check: another model agreed
Report a problem with this finding
Typo Page 1: "The NB distribution has parameters, which are uniquely determined by mean μ and variance σ2" probably lacks the word "two" (page 2 has "which has two parameters").
- p. 1 Background, right column
The NB distribution has parameters, which are
Judge's note: Checked the page 1 image: the sentence is printed as quoted. It follows "The Poisson distribution has a single parameter" and is followed by "to estimate both parameters", and page 2 says "which has two parameters", so a number word was probably dropped. The sentence is still grammatical and readable as printed, so this is a minor point.
Second check: none (judge only)
Report a problem with this finding
Typo Page 1 affiliation footnote prints "European Molecular Biology Laboratory, Mayerhofstraße 1, 69117 Heidelberg"; the street is believed to be Meyerhofstraße.
- p. 1 Affiliation footnote
European Molecular Biology Laboratory, Mayerhofstraße 1, 69117 Heidelberg
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that EMBL Heidelberg's address is Meyerhofstraße 1. The page 1 image shows "Mayerhofstraße 1" as quoted. The paper gives the address only once, so the spelling cannot be checked from the files.
Second check: none (judge only)
Report a problem with this finding
Typo Section name given as "analyses such as shown in Section Application" and, a few lines later, "the analyses shown in Section Applications"; the section is headed "Applications".
- p. 11 The R package DESeq
analyses such as shown in Section Application with
- p. 11 The R package DESeq
such as the analyses shown in Section Applications
- p. 11 Materials and methods
analyses such as shown in Section Application
- p. 6 Comparison with edgeR
tag-wise dispersion mode
Judge's note: Page 11 image shows both forms in the same paragraph. The Background (text line 89) also uses "Section Applications". The other points in this cluster are not defects I would report: "tag-wise" versus "tagwise" on page 6 is a hyphenation variant, and "the Discussion" on page 1 is a loose reference under a "Results and Discussion" heading.
Second check: none (judge only)
Report a problem with this finding
Typo Page 3, after Equation (6): "the sum runs over these replicates. the functions vρ, we first calculate sample variances on the common scale" has lost the start of the sentence (probably "To estimate").
- p. 3 Fitting, text after Equation (6)
the sum runs over these replicates. the functions v
- p. 3 after Eq. (6)
the sum runs over these replicates. the functions vr, we
- p. 3 after Eq. 6
the sum runs over these replicates. the functions
Judge's note: The page 3 image shows the same text: a lower-case "the functions vρ, we first calculate" directly after a full stop. The parallel sentence before Equation (6) begins "To estimate qiρ, we use", so "To estimate" is the evident missing text.
Second check: none (judge only)
Report a problem with this finding
Typo Equations (12) and (13) print the size factor as "s_j" without the estimate hat, while Equations (6), (7), (8) and (14) print ŝ_j.
- p. 3 Equation (12)
kij / sj
- p. 4 Equation (13)
which accounts for the fact that the null hypothesis
Judge's note: Checked the page 3 and page 4 images. Equation (12) has "k_ij / s_j" and Equation (13) has "s_j q̂_i0", both without a hat. Equation (14), the companion of (13), has ŝ_j, as do Equations (5) to (8). This is a notation inconsistency with no effect on meaning.
Second check: none (judge only)
Report a problem with this finding
Typo Page 10: "the method of Oshlack and Robinson [13]" reverses the author order of reference 13 ("Robinson MD, Oshlack A"), which page 3 cites as "Robinson and Oshlack [13]".
- p. 3 Fitting, after Equation (5)
Robinson and Oshlack [13] suggested a similar method
- p. 10 Sharing of information between genes
method of Oshlack and Robinson [13]
- p. 11 Reference 13
Robinson MD, Oshlack A: A scaling normalization method
- p. 3 Estimating size factors
Robinson and Oshlack [13]
Judge's note: Checked pages 3, 10 and 11. The reference list gives "Robinson MD, Oshlack A"; page 3 follows that order and page 10 reverses it. One reviewer also noted "edgeR 1, 127 hits" on page 6: the image does show a space inside the number 1,127, which is cosmetic. The "Figure 2edgeR" point is handled in C015.
Second check: none (judge only)
Report a problem with this finding
Typo Page 6: "as we have seen in Figure 2edgeR is anti-conservative" is missing a comma and space between "Figure 2" and "edgeR".
- p. 6 Comparison with edgeR
seen in Figure 2edgeR is anti-conservative
- p. 6 Comparison with edgeR
as we have seen in Figure 2edgeR is anti-conservative
Judge's note: The page 6 image and crop show "Figure 2" running directly into the italic "edgeR" with no comma or space, so this is in the print and not only in the text layer.
Second check: none (judge only)
Report a problem with this finding
Typo Stray space inside a number: "edgeR 1, 127 hits" should read "1,127 hits".
- p. 6 Comparison with edgeR
edgeR 1, 127 hits
Judge's note: Page 6 image shows a clear gap after the thousands comma in "edgeR 1, 127 hits"; the text layer has the same space. Other numbers in the paper (e.g. 17,605) have no space.
Second check: none (judge only)
Report a problem with this finding
Typo The list of differences from edgeR runs "First, we discovered...", "Third, we have suggested...", "Fourth, our approach provides..." with no item labelled "Second".
- p. 9 Sharing of information between genes
First, we discovered that the use of total read counts
- p. 10 Sharing of information between genes
Third, we have suggested a simple and robust way
- p. 10 Conclusions, left column, second full paragraph (where "Second" would be expected)
For small numbers of replicates as often encountered
- p. 10 Conclusions, right column
Fourth, our approach provides useful diagnostics.
Judge's note: Page 9 image has "First, we discovered that the use of total read counts"; page 10 image has "Third, we have suggested a simple and robust way" and "Fourth, our approach provides useful diagnostics." The paragraph "For small numbers of replicates as often encountered" in between has no marker. A text search finds "Second" only on earlier pages, in unrelated lists.
Second check: none (judge only)
Report a problem with this finding
Typo Figure 6 caption misspells "fourth" twice: "(third and forth column)" and "The forth column shows".
- p. 9 Figure 6 caption
(third and forth column)
- p. 9 Figure 6 caption
The forth column
- p. 9 Figure 6 legend
third and forth column
Judge's note: Page 9 image shows both instances in the Figure 6 caption.
Second check: none (judge only)
Report a problem with this finding
Typo Subject-verb disagreement: "the negative binomial distribution is a generalization that allow for the variance to be larger" should read "allows".
- p. 9 Choice of distribution, last paragraph
binomial distribution is a generalization that allow for
Judge's note: Page 9 image, right column, shows "a generalization that allow for the variance to be larger".
Second check: none (judge only)
Report a problem with this finding
Typo Reference 24 editor list is garbled: "Edited by: Gentleman R, Carey V, Dudoit S, R Irizarry WH."
- p. 11 Reference 24
Gentleman R, Carey V, Dudoit S, R Irizarry WH
Judge's note: Page 11 image shows "Gentleman R, Carey V, Dudoit S, R Irizarry WH". The last item breaks the Surname-Initial pattern of the rest of the list and appears to merge two names; reference 28 in the same list has "Huber W" and "Irizarry R", so the likely intended reading is "Irizarry R, Huber W". The garbling is visible without outside knowledge.
Second check: none (judge only)