RSEM: Accurate Transcript Quantification from RNA-Seq Data with or without a Reference Genome
BMC Bioinformatics, 2011 · paper · Biomedicine
Presents RSEM, software that estimates gene and isoform abundances from RNA-Seq data without needing a reference genome. The authors report performance superior or comparable to genome-based methods and use it to guide experimental design.
Checked: PDF from europepmc.org, fetched 4 Oct 2026. Page numbers are those of that PDF.
5 findings
Reference or label Page 5: "the original implementation of RSEM (v0.6) [32]" cites reference 32 (Li J, Jiang H, Wong WH: Modeling non-uniformity in short-read rates in RNA-Seq data), which is not the RSEM paper; the original RSEM is reference 7 (Li B, Ruotti V, Stewart RM, Thomson JA, Dewey CN).
- p. 5 Comparison to related tools
the original implementation of RSEM (v0.6) [32]
- p. 15 Reference 32
Li J, Jiang H, Wong WH: Modeling non-uniformity in short-read rates in RNA-Seq data. Genome Biology 2010, 11(5):R50
- p. 15 Reference 7
Li B, Ruotti V, Stewart RM, Thomson JA, Dewey CN: RNA-Seq gene expression estimation with read mapping uncertainty
- p. 5 Results and Discussion, Comparison to related tools
RSEM (v0.6) [32]
Judge's note: Read the page 5 crop and the reference list on page 15. Reference 32 is a Genome Biology 2010 paper by other authors on non-uniform read rates. The paper cites the original RSEM method as [7] everywhere else (for example 'our original statistical model [7]' on page 12, and pages 2 and 4). [32] is cited nowhere else in the text. Settled from the files; the evident fix is [7].
Second check: another model agreed
Report a problem with this finding
Reference or label Figure 3 caption has PE and SE swapped: "estimates from PE data are shown in (A) and (B), respectively. Global isoform and gene estimate error distributions for SE data are shown in (C) and (D)"; panels A and B are the SE results and C and D the PE results.
- p. 7 Figure 3 caption
estimates from PE data are shown in (A) and (B), respectively. Global isoform and gene estimate error distributions for SE data are shown in (C) and (D)
- p. 6 Simulated data, paragraph on Figure 3
RSEM v0.6 and rQuant were only run on the SE data, as they do not handle PE data.
- p. 7 Figure 3 caption
from PE data are shown in (A) and (B), respectively. Global isoform and gene estimate error distributions for SE data are shown in (C) and (D)
- p. 7 Table 1, PE columns
RSEM v0.6
Judge's note: Looked at page 7. The legends of panels A and B list five methods including rQuant and RSEM v0.6; C and D list only RSEM, IsoEM and Cufflinks. Page 6 says 'RSEM v0.6 and rQuant were only run on the SE data, as they do not handle PE data', and Table 1 has no PE values for them. The Cufflinks curves agree: in C and D about 85-90% of items exceed 10% error, matching Table 1's PE 10% EF of 89.3/86.4, while in A and B it is about 50%, matching the SE values 46.6/53.3. The isoform/gene order (A and C isoforms, B and D genes) matches the y-axis labels, so only the PE/SE assignment is wrong. No statement elsewhere explains it.
Second check: another model agreed
Report a problem with this finding
Wording Mean isoforms per gene for the human RefSeq annotation is given as 1.7 on page 8 ("compared to 1.7 for all genes") and 1.6 on page 10 ("has more isoforms per gene on average (1.6)").
- p. 8 MAQC data, last paragraph
number of isoforms per gene in this set is 1.1, compared to 1.7 for all genes
- p. 10 Paired vs. single end reads
has more isoforms per gene on average (1.6) than the mouse RefSeq annotation (1.2)
- p. 8 MAQC data, paragraph on lack of distinction between methods
to 1.7 for all genes
- p. 10 Paired vs. single end reads, right column
has more isoforms per gene on average (1.6) than the
Only the authors can say what is right here.
Judge's note: Read pages 8 and 10. Page 8 compares the qRT-PCR subset with 'all genes' of the 19,005-gene RefSeq set; page 10 describes 'the human RefSeq annotation'. Methods (Reference sequences, pages 13-14) describe one human RefSeq set (hg18, filtered for non-coding genes and non-standard chromosomes) and give no second set or definition that would explain two means. The difference is small and does not affect any conclusion, hence S3. Only the authors can say which value is right or whether the sets differ.
Second check: another model agreed
Report a problem with this finding
Wording Page 8 says "After filtering for the qRT-PCR genes that were consistent in their annotation with RefSeq and had non-zero abundance (see Methods), only 716 could be used for correlation analysis", but Methods (page 14) give 716 as the count after the annotation filter alone: "This filtering resulted in a set of 716 genes, 656 and 618 of which were detected in UHR and HBR, respectively."
- p. 8 MAQC data, last paragraph
consistent in their annotation with RefSeq and had non-zero abundance (see Methods), only 716 could be used for correlation analysis
- p. 14 Methods, MAQC validation
This filtering resulted in a set of 716 genes, 656 and 618 of which were detected in UHR and HBR, respectively.
- p. 8 MAQC data, right column
zero abundance (see Methods), only 716 could be used
- p. 14 Methods, MAQC validation
resulted in a set of 716 genes, 656 and 618 of which
Only the authors can say what is right here.
Judge's note: Read page 8 and the page 14 crop. In Methods, 716 follows from the accession-consistency filter only; detection reduces it to 656 (UHR) and 618 (HBR). Page 8 also says correlations used only genes with non-zero abundance by qRT-PCR and all methods, so the number used per sample is at most 656 or 618, not 716. The Results sentence attaches the non-zero condition to the wrong count. Nothing elsewhere reconciles it (716 is not described as a union across samples).
Second check: another model agreed
Report a problem with this finding
Wording Table 2 footnote describes the values as "Pearson r2 of log-transformed abundance values" (superscript 2 on the page), while the text says "we calculated the Pearson correlation of the logarithm of the abundance values" and the table title says 'Correlation'.
- p. 8 MAQC data
we calculated the Pearson correlation of the logarithm of the abundance values
- p. 8 Table 2 footnote
Correlation values (Pearson r2 of log-transformed abundance values)
- p. 8 MAQC data, left column
lated the Pearson correlation of the logarithm of the
- p. 8 Table 2 caption
Pearson r2 of log-transformed abundance values
Only the authors can say what is right here.
Judge's note: Looked at page 8: the footnote prints an italic r with a superscript 2; the body text and table title say Pearson correlation with no mention of squaring. r and r squared are different quantities (0.69 as r would be about 0.48 as r squared), and the paper does not say which is tabulated. The 'four methods' count in the footnote is fine (RSEM, IsoEM, Cufflinks, rQuant, with two bias-correction variants).
Second check: another model agreed
Report a problem with this finding
12 minor (typos and the like; most checked by the judge only)
Typo Abstract: "We present RSEM, an user-friendly software package" should read "a user-friendly".
- p. 1 Abstract, Results
We present RSEM, an user-friendly software package
- p. 2 Background, first line of page
present a user-friendly software package
- p. 1 Abstract, Results
an user-friendly software package
Judge's note: Read the abstract crop and page 2. The abstract prints "an user-friendly"; the Background on page 2 prints "present a user-friendly software package" for the same phrase. Plain grammatical slip.
Second check: none (judge only)
Report a problem with this finding
Typo Mismatched quotation marks: "("multi-reads”)" on page 2, and "("gene multireads”)" and "“rescue"-like strategy" on page 6 mix straight and curly quotes within one quoted term.
- p. 2 Related work
("multi-reads”)
- p. 6 Simulated data
("gene multireads”)
- p. 6 Simulated data
“rescue"-like strategy
Judge's note: Visible in the page 6 crop: a straight opening quote before 'gene' with a curly closing quote after 'multireads', and a curly opening quote before 'rescue' with a straight closing quote. The same pattern is on page 2 for 'multi-reads'. Not a text-layer artefact; purely cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Page 4: "unfiltered fragments that are derived from a isoform or gene" should read "an isoform".
- p. 4 Read mapping and abundance estimation, output description
unfiltered fragments that are derived from a isoform or gene
- p. 4 Implementation, abundance estimates
derived from a isoform or gene
Judge's note: Confirmed on the page 4 crop; the words are intact on the page.
Second check: none (judge only)
Report a problem with this finding
Typo Methods subsection heading on PDF page 12 reads "Expectation-Maximation"; it should be "Expectation-Maximization", as on page 4 ("Expectation-Maximiza-tion (EM) algorithm").
- p. 12 Methods subsection heading
Expectation-Maximation
- p. 4 Implementation
Expectation-Maximiza-
Judge's note: Viewed both crops. The heading on page 12 is a single line printed "Expectation-Maximation", so line-break hyphenation does not explain it. Page 4 spells the term correctly across a line break. Plain misspelling.
Second check: none (judge only)
Report a problem with this finding
Typo Page 9: "(4) 40 million 35 base SE reads" lacks the comma used in the other three configurations ("20 million, 35 base SE reads" etc.).
- p. 9 Paired vs. single end reads, configuration list
(4) 40 million 35 base SE reads
Judge's note: Confirmed on the page 9 crop. Items (1) to (3) have a comma after 'million'; item (4) does not. Trivial punctuation inconsistency with no effect on meaning.
Second check: none (judge only)
Report a problem with this finding
Typo Table 4 header "Throughput (MB)" gives megabases as 'MB', which conventionally means megabytes; 'Mb' would be the usual abbreviation.
- p. 10 Table 4 header
Throughput (MB)
Judge's note: Looked at Table 4 on page 10. The column is read count times read length in millions of bases (20 million x 35 = 700; PE 20 million x 2 x 35 = 1400), so megabases are meant. 'MB' for megabases is loose but common usage and no reader would be misled.
Second check: none (judge only)
Report a problem with this finding
Typo Reference 31 prints the last author as "Zahler AM, Haussler , David :" instead of the surname-plus-initials form used for every other author ("Haussler D:").
- p. 15 Reference 31
Zahler AM, Haussler , David :
- p. 15 Reference 31
Haussler , David :
Judge's note: Viewed the full page 15 image; the entry reads "...Zahler AM, Haussler , David : The Human Genome Browser at UCSC." with stray spaces before the comma and colon and the first name spelled out. Reference 26 on the same page prints the same author as "Haussler D", so the intended form is evident from the paper itself. The text layer agrees with the image.
Second check: none (judge only)
Report a problem with this finding
Typo Reference 36 (MAQC consortium) lists an author as "Puri RK, Schrf U, Thierry-Mieg J"; "Schrf" is probably a misspelling of "Scherf".
- p. 15 Reference 36
Puri RK, Schrf U, Thierry-Mieg J
- p. 15 Reference 36 author list
Schrf U
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the MAQC paper's author is Scherf U. The page 15 image and the text layer both print "Schrf U", so the quote is as published. The correct spelling cannot be verified from the files provided, although a surname with no vowel makes a dropped letter likely.
Second check: none (judge only)
Report a problem with this finding
Typo The reference list spells an author "Rätsch G" in reference 2 ("Bohnert R, Rätsch G") and "Ratsch G" in reference 23 ("Schneeberger K, Ratsch G").
- p. 15 Reference 2
Bohnert R, Rätsch G
- p. 15 Reference 23
Schneeberger K, Ratsch G
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the two entries name the same person (Gunnar Rätsch). The page 15 image shows the umlaut in reference 2 and none in reference 23, and the text layer agrees. This is a minor diacritic inconsistency within the list, possibly inherited from the source record for reference 23.
Second check: none (judge only)
Report a problem with this finding
Typo References 2, 26, 27 and 34 have a stray doubled comma after the year: "Nucleic Acids Research 2010, , 38 Web Server: W348-51.", "2011, , 39 Database: D876-82.", "2011, , 39 Database: D800-6.", "2009, , 37 Database: D32-6."
- p. 15 Reference 2
Nucleic Acids Research 2010, , 38 Web Server: W348-51.
- p. 15 Reference 26
Nucleic Acids Research 2011, , 39 Database: D876-82.
- p. 15 Reference 27
Nucleic Acids Research 2011, , 39 Database: D800-6.
- p. 15 Reference 34
Nucleic Acids Research 2009, , 37 Database: D32-6.
Judge's note: Viewed the full page 15 image; all four entries show ", ," between the year and the volume, and the text layer agrees. Other entries use the form "2010, 38(14):4570-8". This is a formatting defect in the special-issue entries, where the volume is also not in bold.
Second check: none (judge only)
Report a problem with this finding
Typo References 8, 21, 29, 32, 38 and 40 end in a doubled full stop: "11(10):R106..", "38(10):e112..", "10(3):R25..", "11(5):R50..", "12(3):R22..", "38(12):e131.."
- p. 15 Reference 8
11(10):R106..
- p. 15 Reference 21
38(10):e112..
- p. 15 Reference 29
10(3):R25..
- p. 15 Reference 32
11(5):R50..
Judge's note: Viewed the page 15 image for references 8, 21, 29, 32 and 38, and the page 16 crop for reference 40; each ends in "..". All six are entries with an article identifier in place of a page range. Entries with page ranges end in a single full stop. The text layer agrees.
Second check: none (judge only)
Report a problem with this finding
Typo Reference 12 prints author initials as "Thompson Da" and reference 36 prints "Fan Xh"; every other author in the list has fully capitalised initials ("Thompson DA", "Fan XH").
- p. 15 Reference 12 author list
Thompson Da
Judge's note: Viewed the full page 15 image; "Thompson Da, Amit I" appears in reference 12 and "Fan Xh, Fang H" in reference 36, and the text layer agrees. The lowercase second initial is inconsistent with the rest of the list, for example "Thomson JA" in reference 7 and "Levin JZ" in reference 12 itself.
Second check: none (judge only)