STAR: Ultrafast Universal RNA-seq Aligner
2013 · paper · Biomedicine
Introduces STAR, software that aligns RNA sequencing reads to a genome using a seed-search, clustering and stitching algorithm. The authors report it maps reads about 450 times faster than other aligners, with improved sensitivity and precision, and validated novel splice junctions experimentally.
Checked: PDF from europepmc.org, fetched 4 Oct 2026. Page numbers are those of that PDF.
7 findings
Substance Abstract: "we experimentally validated 1960 novel intergenic splice junctions"; Section 3.4 says "Experimental validation was carried out on 1920 novel splice junctions" and "We selected 1920 intergenic and antisense splice junctions", and the tested-junction counts in Table 2 sum to 1920.
- p. 1 Abstract, Results
we experimentally validated 1960
- p. 5 Section 3.4, paragraph 3
Experimental validation was carried out on 1920 novel splice
- p. 5 Section 3.4, paragraph 4
We selected 1920 intergenic and antisense splice junctions
- p. 6 Table 2, 'Number of tested junctions' columns
192
Judge's note: Read the abstract on the page 1 image (1960), Section 3.4 on the page 5 image (1920 twice) and Table 2 on the page 6 image. Each cell line has tested counts 192, 192, 96, 96, 96, 96, 96, 96 = 960, so 1920 in total. No other junction set or exclusion step is mentioned anywhere in the article, so nothing explains 1960. The abstract figure is the odd one out. 1920 is also the number tested, not the number validated.
Second check: another model agreed · checked by hand
Report a problem with this finding
Reference or label Introduction cites "Han et al., 2011" in a list of "Various sequence alignment algorithms"; the reference list entry is "Han,J. et al. (2011) Pre-mRNA splicing: where and when in the nucleus. Trends Cell. Biol., 21, 336–343."
- p. 1 Introduction, paragraph 2
Grant et al., 2011; Han et al., 2011; Trapnell et al.
- p. 7 References
Han,J. et al. (2011) Pre-mRNA splicing: where and when in the nucleus. Trends
- p. 1 Introduction, paragraph 2
Grant et al., 2011; Han et al., 2011; Trapnell et al.,
- p. 1 Introduction
Han et al., 2011
Only the authors can say what is right here.
Judge's note: Page 1 image: the citation sits in the list introduced by "Various sequence alignment algorithms have been recently developed to tackle these challenges". Page 7 image: the only Han entry is a Trends Cell Biol. review whose title is about pre-mRNA splicing in the nucleus, not an alignment method. The other entries cited in that list (Au, De Bona, Grant, Trapnell, Wang, Wu and Nacu, Zhang) have titles about alignment or splice junction detection. The mismatch is visible from the title alone; which work was intended is for the authors to say.
Second check: another model agreed
Report a problem with this finding
Reference or label Section 3.2: "The percentage of the unannotated among all detected junctions is plotted in Figure 3b"; the Fig. 3b y-axis reads "% of annotated junctions" and the caption says "(b) percentage of detected junctions that are annotated".
- p. 4 Section 3.2, text below Fig. 3, left column
The percentage of the unannotated among all detected junctions is plotted in Figure 3b
- p. 4 Fig. 3 caption, (b)
(b) percentage of detected junctions that are annotated
- p. 4 Fig. 3b y-axis label
% of annotated junctions
- p. 4 Section 3.2 text below Fig. 3, left column
The percentage of the unannotated among all
Judge's note: Page 4 image: panel (b) has the y-axis label "% of annotated junctions", with curves rising from about 50-65% towards 100% as the threshold increases, which is the annotated share. The caption says annotated. The next sentence of the text itself speaks of "the proportion of annotated among all detected junctions". The two quantities are complements, so the data are unaffected, but the sentence names the wrong one for the panel.
Second check: another model agreed
Report a problem with this finding
Wording The discussion says "STAR exhibits better alignment precision and sensitivity than other RNA-seq aligners for both experimental and simulated data" (abstract: "improving alignment sensitivity and precision"), but Section 3.2 reports STAR "achieving the second in class pseudo-sensitivity (Fig. 3f). GSNAP exhibits the highest pseudo-sensitivity", and Fig. 2 shows GSNAP reaching a higher true-positive rate than STAR on simulated data.
- p. 6 Discussion, paragraph 2
STAR exhibits better alignment precision and sensitivity
- p. 4 Section 3.2, right column
ing the second in class pseudo-sensitivity (Fig. 3f). GSNAP
- p. 4 Section 3.2, right column
exhibits the highest pseudo-sensitivity at the cost of a high
- p. 3 Fig. 2, GSNAP and STAR end points
True Positive Rate %
Only the authors can say what is right here.
Judge's note: Checked the discussion on page 6, Section 3.2 and Fig. 3f on page 4, Fig. 2 on page 3 and the abstract on page 1. Fig. 3f: GSNAP tops out near 94%, STAR near 82%. Fig. 2: STAR's end point is about 89% TPR at about 1% FPR, GSNAP's about 94-95% at about 3.7% (read from the plots, approximate). STAR's curve is above the others at a matched false-positive rate, which may be what the authors mean, but the discussion claim is unqualified and the results text itself ranks STAR second. An overstatement in wording; the data are not in question.
Second check: another model agreed
Report a problem with this finding
Wording Section 3.4: of junctions supported by at least five RNA-seq reads, "~82–89% (H1ES) and 84–95% (HUVEC) were corroborated"; Table 2 H1ES rows for 5 or more reads are 82.3, 79.2, 81.3, 87.5 and 88.5, so the lowest is 79.2 (row 6–7), below the stated 82.
- p. 5 Section 3.4, last paragraph
82–89% (H1ES) and 84–95% (HUVEC) were corroborated
- p. 6 Table 2, H1ES, row 6–7
79.2
- p. 5 Section 3.4
82–89% (H1ES) and 84–95% (HUVEC)
- p. 6 Table 2, H1ES column
6–7 96 79.2
Only the authors can say what is right here.
Judge's note: Page 5 image shows "~82–89% (H1ES) and 84–95% (HUVEC)" (the tilde is lost in the text layer). Page 6 image, Table 2 H1ES: 5 reads 82.3, 6-7 reads 79.2, 8-11 reads 81.3, 12-24 reads 87.5, >=25 reads 88.5. HUVEC rows for 5 or more reads are 84.4, 84.4, 86.5, 94.8, 90.6, which matches 84-95%. The H1ES range matches the first and last rows, not the minimum and maximum. The tilde marks it as approximate, but 79.2 does not round to 82. No statement in the article gives another grouping. Either the text range or the 79.2 entry needs checking by the authors. Rated S3 because the range is given as approximate and the gap is three points.
Second check: another model agreed
Report a problem with this finding
Wording Section 3.3: "Like all other aligners, with the exception of RUM, the amount of RAM used by STAR does not increase significantly with the number of threads"; Table 1 gives TopHat2 peak RAM of 4.1 GB at 6 threads and 11.3 GB at 12 threads.
- p. 5 Section 3.3, paragraph 3
Like all other aligners, with the exception of RUM
- p. 5 Table 1, TopHat2 row, Peak physical RAM
11.3
- p. 5 Section 3.3, paragraph 3
Like all other aligners, with the exception of RUM,
- p. 5 Table 1
TopHat2
Only the authors can say what is right here.
Judge's note: Page 5 image, Table 1 peak physical RAM (6 / 12 threads): STAR 27.0 / 28.4, STAR sparse 15.6 / 16.0, TopHat2 4.1 / 11.3, RUM 26.9 / 53.8, MapSplice 3.3 / 3.3, GSNAP 25.9 / 27.0. TopHat2 rises 2.76-fold, more in relative terms than RUM (2.0-fold), yet RUM is the only exception the text names. The absolute rise for TopHat2 (7.2 GB) is smaller than RUM's (26.9 GB), and a table entry could be mistyped, so the authors should say which is right. As printed, text and table disagree.
Second check: another model agreed
Report a problem with this finding
Wording Section 3.3 states "Like all other aligners, with the exception of RUM, the amount of RAM used by STAR does not increase significantly with the number of threads", but Table 1 shows TopHat2 peak RAM rising from 4.1 GB (6 threads) to 11.3 GB (12 threads), a larger relative increase than RUM (26.9 to 53.8 GB).
- p. 5 Section 3.3
with the exception of RUM, the amount of RAM used by STAR does not increase signifi-
- p. 5 Table 1, TopHat2 row
4.1
Only the authors can say what is right here.
Judge's note: Page 5 image, Table 1 peak physical RAM columns: STAR 27.0/28.4, STAR sparse 15.6/16.0, TopHat2 4.1/11.3, RUM 26.9/53.8, MapSplice 3.3/3.3, GSNAP 25.9/27.0. TopHat2 grows by a factor of about 2.8 against 2.0 for RUM, yet only RUM is named as the exception. No footnote or text explains the TopHat2 values. The statement about STAR itself holds (27.0 to 28.4).
Second check: another model agreed
Report a problem with this finding
7 minor (typos and the like; most checked by the judge only)
Wording Abstract: "novel intergenic splice junctions with an 80–90% success rate"; Section 3.4 describes the set as "intergenic and antisense splice junctions" and Table 2 validation rates run from 72.4% to 94.8%.
- p. 1 Abstract, Results
novel intergenic splice junctions with an 80–90% success rate
- p. 5 Section 3.4, paragraph 4
intergenic and antisense splice junctions from
- p. 1 Abstract
with an 80–90% success rate
- p. 6 Table 2
94.8
Judge's note: Page 1 image for the abstract, page 5 image for Section 3.4, page 6 image for Table 2. The body says only junctions mapped to "intergenic or antisense loci" were chosen, so "intergenic" in the abstract covers part of the set. Table 2 rates are 72.4-88.5 (H1ES) and 74.0-94.8 (HUVEC); the body gives ~82-89% and 84-95% for junctions with at least five reads and 72%/74% for two reads. "80-90%" is a loose summary of these. Abstract shorthand, either form could be defended, so observation only. The 1960 vs 1920 count is handled in C001.
Second check: another model agreed
Report a problem with this finding
Typo Abstract: "To align our large (>80 billon reads) ENCODE Transcriptome RNA-seq dataset"; "billon" should be "billion".
- p. 1 Abstract, Results
(>80 billon reads)
Judge's note: The page 1 image shows "(>80 billon reads)" in the abstract and "(>80 billion Illumina reads)" in the last paragraph of the Introduction. The misspelling is printed. The text layer renders ">" as "4" ("480"), which is an extraction artefact and not part of the defect.
Second check: none (judge only)
Report a problem with this finding
Typo Introduction citation list: "De Bona, et al., 2008" has a comma between the author name and "et al.", unlike the other citations ("Au et al., 2010", "Grant et al., 2011").
- p. 1 Introduction, paragraph 2
De Bona,
- p. 1 Introduction, citation list
De Bona, et al., 2008
Judge's note: Page 1 image, right column: the line ends "De Bona," and the next begins "et al., 2008". The comma is printed; the line break does not create it. All other citations in the same list have no comma before "et al.".
Second check: none (judge only)
Report a problem with this finding
Typo MMP is expanded as "Maximal Mappable Prefix (MMP)" in Section 2.1 and as "Maximum Mappable Prefix" in the Fig. 1 caption.
- p. 2 Section 2.1, first sentence
search for a Maximal Mappable Prefix (MMP)
- p. 2 Fig. 1 caption
Schematic representation of the Maximum Mappable Prefix
- p. 2 Section 2.1
Maximal Mappable Prefix (MMP)
- p. 2 Fig. 1 caption
Maximum Mappable Prefix
Judge's note: Page 2 image: Section 2.1 first sentence reads "search for a Maximal Mappable Prefix (MMP)"; the Fig. 1 caption reads "Schematic representation of the Maximum Mappable Prefix search". Two expansions of one abbreviation on the same page. The meaning does not change, so this is a consistency point only.
Second check: none (judge only)
Report a problem with this finding
Typo Section 2.1: "The MMP in STAR search is implemented through uncompressed suffix arrays (SAs)"; the words are out of order (evidently "The MMP search in STAR is implemented").
- p. 2 Section 2.1, right column, first line
The MMP in STAR search is implemented through
Judge's note: Page 2 image, first line of the right column: the printed order is "The MMP in STAR search is implemented through". Not an extraction reordering. Later on the same page the paper writes "the MMP search, implemented in STAR".
Second check: none (judge only)
Report a problem with this finding
Typo Section 2.1: "Mummer (Delcher et al., 1999, 2002; Kurtz et al.) and MAUVE"; the Kurtz citation has no year (the reference list has "Kurtz,S. et al. (2004)").
- p. 2 Section 2.1, paragraph 1
1999, 2002; Kurtz et al.) and MAUVE
- p. 7 References
Kurtz,S. et al. (2004) Versatile and open software for comparing large genomes.
- p. 2 Section 2.1
Kurtz et al.)
- p. 2 Section 2.1
Kurtz et al.) and MAUVE
Judge's note: Page 2 image shows "Kurtz et al.)" with no year; every other citation in the article carries one. Page 7 image shows the entry Kurtz,S. et al. (2004), which is also listed before Kent,W.S. (2002), out of alphabetical order.
Second check: none (judge only)
Report a problem with this finding
Typo Reference list: "Kent,W.S. (2002) BLAT–the BLAST-like alignment tool. Genome Res., 12, 656–664." is placed after the Kurtz entry, out of alphabetical order, and gives the initials as W.S. where the BLAT author is reportedly W.J. Kent.
- p. 7 References
Kent,W.S. (2002) BLAT–the BLAST-like alignment tool.
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the author of the BLAT paper is W. James Kent (initials W.J.); this cannot be checked from the files. The page 7 image shows the entry printed as "Kent,W.S." and shows it after "Kurtz,S. et al. (2004)" and before "Landt,S.G.", which is out of order in an otherwise alphabetical list; the ordering point is verifiable on the page and is cosmetic.
Second check: none (judge only)