Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2013 Nov 18;8(11):e79448.
doi: 10.1371/journal.pone.0079448. eCollection 2013.

rSeqDiff: Detecting Differential Isoform Expression From RNA-Seq Data Using Hierarchical Likelihood Ratio Test

Affiliations
Free PMC article

rSeqDiff: Detecting Differential Isoform Expression From RNA-Seq Data Using Hierarchical Likelihood Ratio Test

Yang Shi et al. PLoS One. .
Free PMC article

Abstract

High-throughput sequencing of transcriptomes (RNA-Seq) has recently become a powerful tool for the study of gene expression. We present rSeqDiff, an efficient algorithm for the detection of differential expression and differential splicing of genes from RNA-Seq experiments across multiple conditions. Unlike existing approaches which detect differential expression of transcripts, our approach considers three cases for each gene: 1) no differential expression, 2) differential expression without differential splicing and 3) differential splicing. We specify statistical models characterizing each of these three cases and use hierarchical likelihood ratio test for model selection. Simulation studies show that our approach achieves good power for detecting differentially expressed or differentially spliced genes. Comparisons with competing methods on two real RNA-Seq datasets demonstrate that our approach provides accurate estimates of isoform abundances and biological meaningful rankings of differentially spliced genes. The proposed approach is implemented as an R package named rSeqDiff.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

Figure 1
Figure 1. Illustration of the three models.
(A) A hypothetical gene with three exons and two isoforms in blue and red, respectively. (B) Three models characterizing three biological situations of the gene expression patterns between two conditions. The numbers of red and blue bars represent the relative abundances of the corresponding isoforms in the two conditions.
Figure 2
Figure 2. Models for estimating the exon inclusion level ψ using the junction reads.
(A) The “exon-exon junction model” used by MATS . Exon 1 and 3 are common exons shared by the two isoforms, and exon 2 is the skipped exon unique for the longer isoform. ψ: exon inclusion level; UJC: number of reads mapped to the upstream junction; DJC: number of reads mapped to the downstream junction; SJC: number of reads mapped to the skipping junction. (B) The “two-isoform model” transformed from (A). The abundances of the longer and shorter isoforms are θ1 and θ2, respectively, which are estimated using the junction read counts (UJC, DJC and SJC).
Figure 3
Figure 3. Comparisons of rSeqDiff, MATS, Cuffdiff 2 and RT-PCR assays.
(A) Scatter plot of the Δψ values estimated by rSeqDiff (using junction reads only) and MATS. (B) Scatter plot of the Δψ values estimated by rSeqDiff (using junction reads only) and RT-PCR. (C) Scatter plot of the Δψ values estimated by rSeqDiff (using all reads) and RT-PCR. (D) Scatter plot of the log2 fold changes of isoform abundances between ESRP1 and EV estimated by rSeqDiff and Cuffdiff 2. Transcripts classified as model 0, model 1 and model 2 are shown in green, blue and red, respectively. The solid line is the regression line. The dashed line is the y = x line, which represents perfect agreement of the two methods. Δψ: difference of exon inclusion level between ESRP1 and EV; PCC: Pearson Correlation Coefficient; SCC: Spearman Correlation Coefficient.
Figure 4
Figure 4. Examples comparing the estimates between rSeqDiff, MATS, Cuffdiff 2 and RT-PCR assays.
(A) ARHGAP17 gene. (B) ATP5J2 gene. (C) CSF1 gene. The figures on the left show the gene structure and the coverage of reads mapped to the gene visualized in CisGenome Browser , where the horizontal tracks in the picture are (from top to bottom): genome coordinates, gene structures where introns are shrunken for better visualization and the coverage of reads mapped to the genes in ESRP1 and EV samples. The table to the right each figure shows the estimates from each method. formula image and formula image: exon inclusion levels in ESRP1 and EV, respectively; Δψ: difference of exon inclusion levels between ESRP1 and EV (formula image).
Figure 5
Figure 5. Examples demonstrating the estimates from rSeqDiff.
(A)-(C) show NRCAM gene. (D)–(F) show BACE1 gene. (G)-(I) show SCIN gene. (A)(D)(G) show the gene structure and coverage of reads mapped to the gene. (B)(E)(H) show enlargement of the parts in the red boxes in (A)(D)(G), respectively, emphasizing the alternative spliced exons. In (B), the red box emphasizes the alternative exon that was validated by RT-PCR assay in , and the two red arrows represent the positions of the primers of RT-PCR . (C)(F)(I) show estimated abundances for each gene and its isoforms by rSeqDiff. Values in the brackets are the 95% confidence intervals for the estimates.

Similar articles

See all similar articles

Cited by 10 articles

See all "Cited by" articles

References

    1. Wang ET, Sandberg R, Luo S, Khrebtukova I, Zhang L, et al. (2008) Alternative isoform regulation in human tissue transcriptomes. Nature 456: 470–476. - PMC - PubMed
    1. Mortazavi A, Williams BA, McCue K, Schaeffer L, Wold B (2008) Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat Methods 5: 621–628. - PubMed
    1. Pan Q, Shai O, Lee LJ, Frey BJ, Blencowe BJ (2008) Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing. Nat Genet 40: 1413–1415. - PubMed
    1. Lopez-Bigas N, Audit B, Ouzounis C, Parra G, Guigo R (2005) Are splicing mutations the most frequent cause of hereditary disease? FEBS Lett 579: 1900–1903. - PubMed
    1. Kim E, Goren A, Ast G (2008) Insights into the connection between cancer and alternative splicing. Trends Genet 24: 7–10. - PubMed

Publication types

MeSH terms

Grant support

YS was supported by the Rackham Centennial Summer Research Fellowship and Summer Internship Funds of Certificate in Public Health Genetics (CPHG) Program at University of Michigan. HJ’s research was supported in part by an NIH grant 5U54CA163059-02 and a GAPPS Grant from the Bill & Melinda Gates Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Feedback