Detection of differential transcript usage (DTU) from RNA-seq data is an important bioinformatic analysis that complements differential gene expression analysis. Here we present a simple workflow using a set of existing R/Bioconductor packages for analysis of DTU. We show how these packages can be used downstream of RNA-seq quantification using the Salmon software package. The entire pipeline is fast, benefiting from inference steps by Salmon to quantify expression at the transcript level. The workflow includes live, runnable code chunks for analysis using DRIMSeq and DEXSeq, as well as for performing two-stage testing of DTU using the stageR package, a statistical framework to screen at the gene level and then confirm which transcripts within the significant genes show evidence of DTU. We evaluate these packages and other related packages on a simulated dataset with parameters estimated from real data.
R version: R version 4.0.3 (2020-10-10)
Bioconductor version: 3.12
Publication DOI: 10.12688/f1000research.15398.3
If you have questions about this workflow or any Bioconductor
software, please post these to the
Bioconductor support site.
If the questions concern a specific package, you can tag the post with
the name of the package, or for general questions about the workflow,
tag the post with
rnaseqdtu. Note the
for crafting an optimal question for the support site.
RNA-seq experiments can be analyzed to detect differences across groups of samples in total gene expression – the total expression produced by all isoforms of a gene – and additionally differences in transcript isoform usage within a gene. If the amount of expression switches among two or more isoforms of a gene, then the total gene expression may not change by a detectable amount, but the differential transcript usage (DTU) is nevertheless biologically relevant. DTU is common when comparing expression across cell types: recent analysis of the Genotype-Tissue Expression Project (GTEx) (GTEx Consortium et al. 2017) dataset demonstrated that half of all expressed genes contained tissue-specific isoforms (Reyes and Huber 2018). DTU may produce functionally different gene products through alternative splicing and changes to the coding sequence of the transcript, and DTU may result in transcripts with different untranslated regions on the 5’ or 3’ end of the transcript, which may affect binding sites of RNA binding proteins. (Reyes and Huber 2018) found that alternative usage of transcription start and termination sites was a more common event than alternative splicing when examining DTU events across tissues in GTEx. Specific patterns of DTU have been identified in a number of diseases, including cancer, retinal diseases, and neurological disorders, among others (Scotti and Swanson 2018). Large-scale analyses of cancer transcriptomic data, comparing tumor to normal samples, have identified that protein domain losses are a common feature of DTU in cancer, including domains involved in protein-protein interactions (Vitting-Seerup and Sandelin 2017; Climente-González et al. 2017).
While many tutorials and workflows in the
Bioconductor project address differential gene expression, there are
fewer workflows for performing a differential transcript usage analysis,
which provides critical and complementary information to a
gene-level analysis. Some of the existing Bioconductor packages and
functions that can be used for statistical analysis of DTU include
DEXSeq (originally designed for differential exon usage) (Anders, Reyes, and Huber 2012),
diffSpliceDGE from the edgeR package (Robinson, McCarthy, and Smyth 2009; McCarthy, Chen, and Smyth 2012),
diffSplice from the limma package (Smyth 2004; Law et al. 2014), and
DRIMSeq (Nowicka and Robinson 2016).
Other Bioconductor packages which are designed around a DTU analysis include
stageR (Van den Berge et al. 2017), SGSeq (Goldstein et al. 2016), and
IsoformSwitchAnalyzeR (Vitting-Seerup and Sandelin 2018).
stageR can be used for post-processing of transcript-
and gene-level p-values from DTU detection methods, and will be discussed in this workflow.
SGSeq can be used to visualize
splice events, and leverages DEXSeq or limma
for differential testing of splice variant usage.
The Bioconductor package IsoformSwitchAnalyzeR is
well documented and the vignette available from the
IsoformSwitchAnalyzeR landing page
can be seen as an alternative to this workflow.
IsoformSwitchAnalyzeR is designed for the downstream analysis of
functional consequences of identified isoform switches.
It allows for import of data from various
quantification methods, including Salmon, and allows for statistical
inference using DRIMSeq. In addition, IsoformSwitchAnalyzeR includes
functions for obtaining the nucleotide and amino acid sequence
consequences of isoform switching, which is not covered in this workflow.
Other packages related to splicing can be found at the BiocViews links for
For more information about the Bioconductor project and its core
infrastructure, please refer to the overview by Huber et al. (2015).
We note that there are numerous other methods for detecting differential transcript usage outside of the Bioconductor project. The DRIMSeq publication is a good reference for these, having descriptions and comparisons with many current methods (Nowicka and Robinson 2016). This workflow will build on the methods and vignettes from three Bioconductor packages: DRIMSeq, DEXSeq, and stageR.
Previously, some of the developers of the Bioconductor packages edgeR and DESeq2 have collaborated to develop the tximport package (Soneson, Love, and Robinson 2016) for summarizing the output of fast transcript-level quantifiers, such as Salmon (Patro et al. 2017), Sailfish (Patro, Mount, and Kingsford 2014), and kallisto (Bray et al. 2016). The tximport package focuses on preparing estimated transcript-level counts, abundances and effective transcript lengths, for gene-level statistical analysis using edgeR (Robinson, McCarthy, and Smyth 2009), DESeq2 (Love, Huber, and Anders 2014), or limma-voom (Law et al. 2014). tximport produces an offset matrix to accompany gene-level counts, that accounts for a number of RNA-seq biases as well as differences in transcript usage among transcripts of different length that would bias an estimator of gene fold change based on the gene-level counts (Trapnell et al. 2013). tximport can alternatively produce a matrix of data that is roughly on the scale of counts, by scaling transcript-per-million (TPM) abundances to add up to the total number of mapped reads. This counts-from-abundance approach directly corrects for technical biases and differential transcript usage across samples, obviating the need for the accompanying offset matrix.
Complementary to an analysis of differential gene expression, one can use tximport to import transcript-level estimated counts, and then pass these counts to packages such as DRIMSeq or DEXSeq for statistical analysis of differential transcript usage. Following a transcript-level analysis, one can aggregate evidence of differential transcript usage to the gene level. The stageR package in Bioconductor provides a statistical framework to screen at the gene-level for differential transcript usage with gene-level adjusted p-values, followed by confirmation of which transcripts within the significant genes show differential usage with transcript-level adjusted p-values (Van den Berge et al. 2017). The method controls the overall false discovery rate (OFDR) (Heller et al. 2009) for such a two-stage procedure, which will be discussed in more detail later in the workflow. We believe that stageR represents a principled approach to analyzing transcript usage changes, as the methods can be evaluated against a target error rate in a manner that mimics how the methods will be used in practice. That is, following rejection of the null hypothesis at the gene-level, investigators would likely desire to know which transcripts within a gene participated in the differential usage.
Here we provide a basic workflow for detecting differential transcript usage using Bioconductor packages, following quantification of transcript abundance using the Salmon method (Figure 1). This workflow includes live, runnable code chunks for analysis using DRIMSeq and DEXSeq, as well as for performing stage-wise testing of differential transcript usage using the stageR package. For the workflow, we use data that is simulated, so that we can also evaluate the performance of methods for differential transcript usage, as well as differential gene and transcript expression. The simulation was constructed using distributional parameters estimated from the GEUVADIS project RNA-seq dataset (Lappalainen et al. 2013) quantified by the recount2 project (Collado-Torres et al. 2017), including the expression levels of the transcripts, the amount of biological variability of gene expression levels across samples, and realistic coverage of reads along the transcript.
Note that a published version of this workflow exists in the Bioconductor channel of the journal F1000Research. The published version additionally contains evaluations of the methods shown here alongside other popular methods for DTU, DGE, and DTE.
First we describe details of the simulated data, which will be used in the following workflow. Understanding the details of the simulation will be useful for assessing the methods in the later sections. All of the code used to simulate RNA-seq experiments and write paired-end reads to FASTQ files can be found at an associated GitHub repository for the simulation code (Love 2018a), and the reads and quantification files can be downloaded from Zenodo (Love 2018b). Salmon (Patro et al. 2017) was used to estimate transcript-level abundances for a single sample (ERR188297) of the GEUVADIS project (Lappalainen et al. 2013), and this was used as a baseline for transcript abundances in the simulation. Transcripts that were associated with estimated counts less than 10 had abundance thresholded to 0, all other transcripts were considered “expressed” (n=46,933). alpine (Love, Hogenesch, and Irizarry 2016) was used to estimate realistic fragment GC bias from 12 samples from the GEUVADIS project, all from the same sequencing center (the first 12 samples from CNAG-CRG in Supplementary Table 2 from Love, Hogenesch, and Irizarry (2016)). DESeq2 (Love, Huber, and Anders 2014) was used to estimate mean and dispersion parameters for a Negative Binomial distribution for gene-level counts for 458 GEUVADIS samples provided by the recount2 project (Collado-Torres et al. 2017). Note that, while gene-level dispersion estimates were used to generate underlying transcript-level counts, additional uncertainty on the transcript-level data is a natural consequence of the simulation, as the transcript-level counts must be estimated (the underlying transcript counts are not provided to the methods).
polyester (Frazee et al. 2015) was used to simulate paired-end RNA-seq reads for two groups of 12 samples each, with realistic fragment GC bias, and with dispersion on transcript-level counts drawn from the joint distribution of mean and dispersion values estimated from the GEUVADIS samples. To compare DRIMSeq and DEXSeq in further detail, we generated an additional simulation in which dispersion parameters were assigned to genes via matching on the gene-level count, and then all transcripts of a gene had counts generated using the same per-gene dispersion. The first sample for group 1 and the first sample for group 2 followed the realistic GC bias profile of the same GEUVADIS sample, and so on for all 12 samples. This pairing of the samples was used to generate balanced data, but not used in the statistical analysis. countsimQC (Soneson and Robinson 2017) was used to examine the properties of the simulation relative to the dataset used for parameter estimation, and the full report can be accessed at the associated GitHub repository for simulation code (Love 2018a).
Differential expression across two groups was generated as follows: The 46,933 expressed transcripts as defined above belonged to 15,017 genes. 70% (n=10,514) of these genes with expressed transcripts were set as null genes, where abundance was not changed across the two groups. For 10% (n=1,501) of genes, all isoforms were differentially expressed at a log fold change between 1 and 2.58 (fold change between 2 and 6). The set of transcripts in these genes was classified as DGE (differential gene expression) by construction, and the expressed transcripts were also DTE (differential transcript expression), but they did not count as DTU (differential transcript usage), as the proportions within the gene remained constant. To simulate balanced differential expression, one of the two groups was randomly chosen to be the baseline, and the other group would have its counts multiplied by the fold change. For 10% (n=1,501) of genes, a single expressed isoform was differentially expressed at a log fold change between 1 and 2.58. This set of transcripts was DTE by construction. If the chosen transcript was the only expressed isoform of a gene, this counted also as DGE and not as DTU, but if there were other isoforms that were expressed, this counted for both DGE and DTU, as the proportion of expression among the isoforms was affected. For 10% (n=1,501) of genes, differential transcript usage was constructed by exchanging the TPM abundance of two expressed isoforms, or, if only one isoform was expressed, exchanging the abundance of the expressed isoform with a non-expressed one. This counted for DTU and DTE, but not for DGE. An MA plot of the simulated transcript abundances for the two groups is shown in Figure 2.
This workflow was designed to work with R 3.5 or higher, and the DRIMSeq, DEXSeq, stageR, and tximport packages for Bioconductor version 3.7 or higher. Bioconductor packages should always be installed following the official instructions. The workflow uses a subset of all genes to speed up the analysis, b