TopHat: Discovering Splice Junctions with RNA-Seq

• p. 1 Abstract, Motivation

  current software for aligning RNA-Seq data to a genome relies on known splice junctions and cannot identify novel ones

• p. 1 Introduction, last paragraph on page

  One recent method for ab initio junction mapping relies on a machine learning strategy to identify junctions. QPALMA

• p. 5 Discussion, first paragraph

  QPALMA can accurately align short reads across junctions without an annotation

Judge's note: Read pages 1, 2 and 5. All three quotes are present. Page 2 says QPALMA "uses a training step that requires a set of known junctions", so the abstract's "relies on known splice junctions" can be read loosely as covering it, but "cannot identify novel ones" does not sit with the paper's own description of QPALMA as an ab initio method. This is a wording inconsistency, and either form could be defended.

Second check: another model agreed

• p. 4 Table 1, 'Total (%)' column

  9275 93

• p. 4 Table 1, footnote

  The simulation sampled a set of transcripts with 9879 true splice junctions.

Judge's note: Read Table 1 and its footnote on the page 4 image. Four of the five 'Total (%)' values are one point below the conventionally rounded value, and all five match truncation. 'Reported (%)' matches FP/(TP+FP) truncated (4.67 printed as 4, 2.80 printed as 2); that denominator is inferred, not stated. The truncation is consistent but undeclared, so this is a convention point and not a misprint.

Second check: another model agreed

• p. 4 Results, right column

  contains 1295 transcripts from mouse chromosome 7. These were generated by the short-read simulator from Maq.

Only the authors can say what is right here.

Judge's note: Read on the page 4 image. The following sentences ("errors in the reads it produces", "We generated simulated sequence from the ASTD transcripts") show that the simulator produced reads from the transcripts. The sentence misdescribes this, probably because some wording was lost. The intended meaning is recoverable from context.

Second check: another model agreed

• p. 4 Results, left column, second paragraph

  A further 3209 (10%) of the missed junctions

• p. 4 Results

  3209 (10%)

• p. 4 Results

  3209 (10%) of the missed junctions

Judge's note: Read on the page 4 image. The denominator is the "30 121 junctions reported by ERANGE and not reported by TopHat" in the preceding sentence. The neighbouring 52% (52.09%) and 15% (15.14%) do not distinguish rounding from truncation. The paper truncates elsewhere (Table 1; 81.16% printed as 81.1%), so 10% fits an undeclared truncation convention and is not a misprint.

Second check: another model agreed

• p. 4 Fig. 4 caption

  but comparatively few junctions in more highly transcribed

• p. 5 Results, right column, Velvet paragraph

  its detection rate decreases as RPKM further increases

Judge's note: Looked at Fig. 4 on the page 4 image. The dashed curve peaks near 0.42 at RPKM of roughly 30–40 and declines gently to about 0.33 at RPKM 140. The Results text on page 5 describes this more accurately ("its detection rate decreases as RPKM further increases"). The caption's "comparatively few" is loose and overstated wording, not a data error, and partly a matter of judgement.

Second check: another model agreed

• p. 5 Results, left column

  shows the distribution of E-values for each sequence’s best BLAST hit

• p. 5 Results, left column, paragraph 'The UCSC gene models…'

  shows the distribution of E-values for each sequence’s best BLAST

• p. 5 Fig. 5 caption

  The BLAT E-value distribution

Judge's note: Read on the page 5 image. The same paragraph, the figure title and the caption all name BLAT; only this sentence says BLAST. A tool-name slip with an evident fix, though it could be loose usage since BLAT is the 'BLAST-like alignment tool'.

Second check: another model agreed

• p. 5 Results, end of left column to top of right column

  One example of a junction from the second category is occurred in the ADP-ribosylation factor Arfgef1

• p. 5 Results, right column

  The junction in Figure 7 skips two of the gene’s 38 exons.

• p. 6 Fig. 7 caption

  This junction skips two exons in the ADP-ribosylation gene Arfgef1.

• p. 5 Results, right column, top

  is occurred in the ADP-ribosylation factor Arfgef1

Only the authors can say what is right here.

Judge's note: Read page 5 and page 6 images. The categories are listed as two known exons / known plus novel / two novel, and the counts follow in reverse order; under either ordering the second is known-plus-novel (6077). The text and the Fig. 7 caption ("This junction skips two exons") describe an exon-skipping event and name no novel exon. In Fig. 7 the junction's ends appear to sit at annotated RefSeq exons, but the resolution does not let me settle that. Only the authors can say which category was meant. This cluster is kept for the category question; the 'is occurred' typo is judged under C017 and the gene description under C018.

Second check: another model agreed

• p. 5 Results, top of right column

  the ADP-ribosylation factor Arfgef1

• p. 7 References, Morinaga et al. (1996)

  guanine nucleotide-exchange protein for ADP ribosylation factor (ARF) 1 and ARF3

• p. 6 Fig. 7 caption

  ADP-ribosylation gene Arfgef1

Only the authors can say what is right here.

Judge's note: Checked page 5 text, the Fig. 7 caption on page 6 and the Morinaga entry on page 7. The paper's own cited reference title describes an exchange protein for ARF, not an ARF itself, and the text and caption use two different descriptions. It is probably shorthand for the gene's full name (ARF guanine nucleotide-exchange factor 1); the authors should confirm the intended wording.

Second check: another model agreed

• p. 5 Results, Fig. 5 paragraph

  each sequence’s best BLAST hit against the GenBank mouse EST database

Judge's note: The page 5 image shows "best BLAST hit" in the same paragraph that says the search used BLAT. The Fig. 5 title and caption use BLAT throughout, and BLAST appears nowhere else in the text. The evident fix is "BLAT".

Second check: another model agreed

• p. 2 Introduction, first paragraph on page

  De Bono et al. report that Vmatch

• p. 1 Introduction, last paragraph

  (De Bona et al., 2008)

• p. 6 References

  De Bona,F. et al. (2008) Optimal spliced alignments of short sequence reads.

• p. 2 Introduction

  De Bono et al. report

Judge's note: Checked the page images for pages 1, 2 and 6. The reference list has no "De Bono" entry, and the sentence is about QPALMA/Vmatch, so it is the same work with the author's name misspelled.

Second check: none (judge only)

• p. 2 Methods, left column, third paragraph

  To generate the island sequences, Tophat invokes the

• p. 3 Methods, left column

  For each splice junction, Tophat searches the IUM reads

• p. 2 Methods, second paragraph

  To generate the island sequences, Tophat invokes

• p. 2 Methods

  Tophat invokes the

Judge's note: Both instances are visible on the page images for pages 2 and 3. This is a cosmetic capitalisation inconsistency.

Second check: none (judge only)

• p. 2 Methods, first paragraph

  sequencing errors than the 3′ end. (Hillier et al., 2008). TopHat

Judge's note: Visible on the page 2 image in the first Methods paragraph. The sentence is closed before its parenthetical citation, which then carries its own full stop. Punctuation only.

Second check: none (judge only)

• p. 2 Methods, first paragraph

  one or more alignment

• p. 5 Results, right column

  around 40% of junction in

• p. 6 Discussion, left column

  non-canonical donor and acceptors sites

Judge's note: All three are present on the page images. Page 5 elsewhere has the correct "donor and acceptor sites", and the Fig. 4 caption has "40% of junctions". Cosmetic grammar slips.

Second check: none (judge only)

• p. 4 Fig. 4 caption

  TopHat detects more than 80% reported by ERANGE

Judge's note: Visible in the Fig. 4 caption on the page 4 image. The next sentence has the full form, "more than 72% of all junctions observed by ERANGE". A missing phrase; the meaning is recoverable.

Second check: none (judge only)

• p. 5 Results, top of right column

  second category is occurred in the ADP-ribosylation factor

Judge's note: The page 5 image shows "is occurred" at the top of the right column, so it is not an extraction artefact.

Second check: none (judge only)

• p. 5 Results, right column, Velvet paragraph

  detects around 40% of junction in

Judge's note: The page 5 image, right column, shows "40% of junction in". The preceding sentence uses "junctions".

Second check: none (judge only)

• p. 6 Funding

  National Institues of Health

Judge's note: The page 6 image, Funding line, shows "National Institues of Health".

Second check: none (judge only)

• p. 7 References, Pozzoli entry

  Pozzoli,U. et. al. (2007)

Judge's note: The page 7 image shows "Pozzoli,U. et. al. (2007)"; the other entries on pages 6 and 7 read "et al.".

Second check: none (judge only)