DEXSeq vs. rMATS: Choosing the Right Tool for Differential Exon Usage and Splicing Analysis

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-20

Key Takeaways

• DEXSeq analyzes differential exon usage by modeling exon counts within a generalized linear model, allowing for the inclusion of complex experimental factors and covariates to detect changes in relative exon abundance.
• rMATS detects specific splicing event types such as skipped exons, alternative 5' and 3' splice sites, mutually exclusive exons, and retained introns by quantifying the percent spliced in (PSI) value for each event.
• The choice between DEXSeq and rMATS hinges on the biological question: DEXSeq is for broad exon usage changes, while rMATS is for identifying specific splicing event alterations with direct mechanistic interpretation via PSI values.
• Both tools are parametric and rely on count-based modeling, but benchmarking studies show no consensus on universal performance, necessitating selection based on experimental design, biological question, and data requirements.
• Integrating DEXSeq and rMATS can provide complementary insights, as demonstrated in disease studies where one tool identified exon usage changes while the other pinpointed specific splicing events like intron retention.
• Annotation quality is critical for both tools; inaccurate or incomplete gene annotations can lead to misassignment of reads and erroneous results, underscoring the need for current and organism-specific annotations.

Direct Answer and Scope

Researchers analyzing RNA sequencing data face a practical decision when investigating transcript-level regulation: whether to use DEXSeq for differential exon usage or rMATS for differential splicing events. Both tools are widely cited and actively maintained, yet they answer different biological questions and require different input data. DEXSeq tests whether individual exons are used differently between conditions within a gene, while rMATS detects specific splicing event types such as skipped exons, alternative 5' splice sites, alternative 3' splice sites, mutually exclusive exons, and retained introns. A review of 22 differential splicing tools categorizes DEXSeq and rMATS by their statistical family and level of analysis, noting that benchmarking studies show no consensus on tool performance and that results vary considerably across different scenarios [<a href="#ref-1">1</a>]. This article provides a decision framework based on experimental design, biological question, data format, and interpretation requirements.

Understanding the Biological Question

Differential Exon Usage Versus Splicing Events

The choice between DEXSeq and rMATS begins with the biological question. Differential exon usage asks whether the relative abundance of specific exons changes between conditions, which can reflect changes in isoform expression, alternative promoter usage, or alternative polyadenylation. Differential splicing events ask whether specific splicing patterns, such as exon inclusion or exclusion, change between conditions.

A study of adipocyte differentiation illustrates why this distinction matters. Researchers used RNA-seq data from a differentiation time course and found that splicing was highly dynamic across differentiation and that the dynamics were impacted by metabolic phenotype. Importantly, there was little overlap between genes that were differentially spliced and those that were differentially expressed, positioning alternative splicing as an independent regulatory mechanism whose impact would be missed when looking at gene expression changes alone [<a href="#ref-2">2</a>]. This finding means that researchers who only examine gene-level expression may miss biologically important splicing changes.

When to Choose DEXSeq

DEXSeq is appropriate when the research question concerns exon-level usage changes without prespecifying the splicing event type. This approach is useful for detecting coordinated changes across gene regions, such as the loss of intron retention across multiple exons of a single gene. In a study of Fuchs endothelial corneal dystrophy, DEXSeq analysis showed reduced expression across FN1 gene regions in patient samples, complementing the splicing event information obtained from rMATS [<a href="#ref-3">3</a>].

DEXSeq is also suitable when the experimental design includes complex factors or covariates. The method models exon counts within a generalized linear model framework, allowing researchers to include batch effects, biological covariates, or interaction terms. This flexibility is valuable for experiments with multiple conditions, time points, or treatment groups.

When to Choose rMATS

rMATS is appropriate when the research question concerns specific splicing event types and when the researcher wants to quantify the percent spliced in (PSI) value for each event. PSI represents the proportion of transcripts that include a particular exon or splice site, providing an interpretable biological metric.

A study of Hub1 overexpression in Saccharomyces cerevisiae used rMATS to detect seven exon skipping events, with DYN2 showing significant differential splicing at an FDR of 0.0481 and a delta PSI of negative 0.036. MaxEntScan analysis confirmed that DYN2's 5' splice site was significantly weaker than canonical yeast splice sites, consistent with Hub1's role in facilitating non-consensus splicing [<a href="#ref-4">4</a>]. This example demonstrates how rMATS output connects directly to mechanistic interpretation.

A study of fat-tail development in sheep used rMATS alongside Whippet to identify differential splicing events between fat-tailed and thin-tailed breeds. The researchers applied stringent filters and only considered events detected by both tools in at least three datasets with the same delta PSI trend. This approach yielded 130 high-confidence skipped exon events belonging to 124 genes [<a href="#ref-5">5</a>]. The use of rMATS in this multi-tool strategy highlights its role in event-level detection.

Statistical Approaches and Methodological Differences

DEXSeq Statistical Framework

DEXSeq operates at the exon level and uses a generalized linear model to test for differential exon usage. The method accounts for the total gene expression and tests whether the proportion of reads mapping to each exon changes between conditions. This approach detects changes in relative exon abundance within a gene, which can arise from alternative splicing, alternative promoter usage, or alternative polyadenylation.

The statistical framework in DEXSeq is parametric and relies on negative binomial distributions to model count data. This approach is consistent with the broader family of count-based differential expression methods used in RNA-seq analysis. The method requires exon-level count data, typically generated by counting reads that overlap each exon of a gene.

rMATS Statistical Framework

rMATS operates at the event level and uses a hierarchical model to compare the inclusion levels of specific splicing events between conditions. The method calculates PSI values for each event and tests whether the PSI differs significantly between conditions. rMATS can detect five types of alternative splicing events: skipped exons, alternative 5' splice sites, alternative 3' splice sites, mutually exclusive exons, and retained introns.

The statistical framework in rMATS is also parametric and models the counts of reads supporting inclusion or exclusion of each event. The method accounts for biological variability between replicates and provides false discovery rate adjusted p-values for each event.

Comparison of Statistical Families

The review of 22 differential splicing tools categorizes them by statistical family, including parametric, non-parametric, and probabilistic approaches. DEXSeq and rMATS both fall within the parametric family, meaning they make distributional assumptions about the data. The review notes that benchmarking studies show no consensus on tool performance, revealing considerable variability across different scenarios [<a href="#ref-1">1</a>]. This finding means that neither tool is universally superior and that the choice should depend on the specific experimental context.

Input Data Requirements

DEXSeq Input Requirements

DEXSeq requires exon-level count data. The typical workflow begins with aligned RNA-seq reads in BAM format. Researchers use the DEXSeq package to construct exon counting bins from a gene annotation file, then count reads overlapping each bin. The counting bins account for the fact that exons can be partially overlapping due to alternative splicing, creating distinct counting units.

The DEXSeq workflow is available through Bioconductor, which provides official package documentation, installation instructions, and reproducible genomic analysis workflows [<a href="#ref-6">6</a>]. Researchers should consult the Bioconductor documentation for the specific version of DEXSeq appropriate to their analysis environment.

rMATS Input Requirements

rMATS requires aligned RNA-seq reads in BAM format along with a gene annotation file in GTF format. The tool uses the annotation to define splicing events and then counts reads that support inclusion or exclusion of each event. rMATS can accept either single-end or paired-end reads, and it requires information about read length and library type.

The rMATS workflow is commonly implemented within larger analysis pipelines. The nf-core documentation describes community pipeline standards for usage, configuration, and reproducible workflow context [<a href="#ref-7">7</a>]. Researchers using rMATS within a pipeline should verify that the pipeline version and configuration match their experimental design.

Annotation Quality Considerations

Both tools depend on the quality and completeness of the gene annotation. Incorrect or incomplete annotations can lead to misassignment of reads to exons or events, producing false results. Researchers should use the most current annotation for their organism of interest and should verify that the annotation version is consistent across all samples in the study.

The NCBI provides official descriptions of sequence resources and analysis services that can help researchers identify appropriate annotations [<a href="#ref-8">8</a>]. Researchers should document the annotation version and source in their analysis records to ensure reproducibility.

Practical Workflow Implementation

Step 1: Define the Biological Question

Before selecting a tool, write a clear statement of the biological question. Determine whether the question concerns specific splicing events, such as exon skipping or intron retention, or whether it concerns broader changes in exon usage across gene regions. This decision determines whether rMATS, DEXSeq, or both tools are appropriate.

Step 2: Assess the Experimental Design

Evaluate the number of biological replicates, the presence of batch effects, and the complexity of the experimental design. DEXSeq can accommodate complex designs with multiple factors and covariates within its generalized linear model framework. rMATS can also handle multi-factor designs, but the interpretation of results becomes more complex with additional factors.

Step 3: Prepare Input Data

Generate aligned reads in BAM format using a splice-aware aligner. Verify that the alignment quality is adequate by checking mapping rates and the distribution of reads across genes. Poor alignment quality will affect both DEXSeq and rMATS results.

For DEXSeq, construct exon counting bins using the DEXSeq package and the gene annotation. For rMATS, prepare the GTF annotation file and verify that it contains the necessary information for event definition.

Step 4: Run Quality Control

Perform quality control on the input data before running either tool. Check for sample swaps, batch effects, and outlier samples. The Galaxy Training Network provides accessible workflow training and analysis tutorials that include quality control steps for RNA-seq data [<a href="#ref-9">9</a>]. The Carpentries lessons provide foundational computing and data skills that support reproducible analysis practices [<a href="#ref-10">10</a>].

Step 5: Run the Analysis

Run DEXSeq or rMATS according to the package documentation. For DEXSeq, follow the Bioconductor workflow for constructing counting bins, counting reads, and testing for differential exon usage [<a href="#ref-6">6</a>]. For rMATS, follow the tool documentation for event definition and statistical testing.

Step 6: Interpret Results

Interpret the results in the context of the biological question. For DEXSeq, examine which exons show differential usage and whether the changes cluster in specific gene regions. For rMATS, examine which event types show differential splicing and whether the delta PSI values are biologically meaningful.

Step 7: Validate Findings

Validate a subset of findings using an independent method. Quantitative PCR can confirm changes in isoform expression or exon inclusion. In the Fuchs endothelial corneal dystrophy study, qPCR validation in an independent cohort demonstrated significantly elevated FN1 expression in patient samples, even though DEXSeq analysis showed reduced expression across FN1 gene regions [<a href="#ref-3">3</a>]. This paradoxical finding highlighted the importance of validation and the complexity of interpreting exon-level results.

At a Glance: Tool Comparison Table

Options and Tradeoffs in Tool Selection

Using Both Tools Together

Several published studies have used both DEXSeq and rMATS in the same analysis to gain complementary information. In the Fuchs endothelial corneal dystrophy study, rMATS identified 486 upregulated and 89 downregulated intron retention events in expansion-positive patients compared to controls. Subsequent analysis with the more stringent IRFinder algorithm revealed 10 upregulated intron retention events across nine genes and 6 downregulated events exclusively localized within FN1. DEXSeq analysis showed reduced expression across FN1 gene regions in patient samples [<a href="#ref-3">3</a>]. The combination of tools provided a more complete picture than either tool alone.

Considering Alternative Tools

The review of 22 differential splicing tools notes that tools are categorized by statistical family and level of analysis, with levels including transcript, exon, and event [<a href="#ref-1">1</a>]. Researchers may consider other tools depending on their specific needs. For example, Whippet was used alongside rMATS in the sheep fat-tail study to increase confidence in detected events [<a href="#ref-5">5</a>]. The choice of additional tools should be based on the specific biological question and the strengths of each tool.

Long-Read Sequencing Considerations

Emerging long-read RNA sequencing technologies are discussed as a complement to short-read methods, promising reduced deconvolution needs and further innovation [<a href="#ref-1">1</a>]. Long-read sequencing can directly capture full-length transcripts, potentially reducing the need for computational isoform deconvolution. Researchers planning new experiments should consider whether long-read sequencing might be appropriate for their biological question.

Observations and Measurements

PSI Values as Interpretable Metrics

The percent spliced in value provides a direct measurement of splicing changes. In the Hub1 overexpression study, rMATS detected seven exon skipping events, with DYN2 showing significant differential splicing at an FDR of 0.0481 and a delta PSI of negative 0.036 [<a href="#ref-4">4</a>]. This delta PSI value indicates a small but statistically significant decrease in exon inclusion. Researchers should consider whether delta PSI values are biologically meaningful in their experimental context.

Splicing as an Independent Regulatory Mechanism

The adipocyte differentiation study found little overlap between genes that were differentially spliced and those that were differentially expressed [<a href="#ref-2">2</a>]. This finding demonstrates that splicing changes can occur independently of expression changes, meaning that researchers who only examine gene expression may miss important regulatory events. The study also integrated splicing results with genome-wide association study data and found that type 2 diabetes associated variants were enriched in regions that were differentially spliced in early differentiation [<a href="#ref-2">2</a>].

Splicing Changes in Disease Contexts

Aberrant alternative splicing has been documented in multiple disease contexts. A study of breast cancer patients found that older adults exhibit a higher frequency of aberrant alternative splicing events compared to younger patients. The study identified 390 high-variability-specific splicing events observed exclusively in older adults patients and used unsupervised clustering to reveal three distinct subtypes with different immune cell infiltration profiles and prognostic outcomes [<a href="#ref-11">11</a>].

A study of Fuchs endothelial corneal dystrophy found that the loss of normal intron retention mediated regulation may contribute to pathological FN1 overexpression. Gene ontology analysis of intron retention associated genes revealed enrichment in RNA splicing and extracellular matrix related pathways, supporting a role for intron retention in disease pathogenesis [<a href="#ref-3">3</a>].

Splicing Regulation by Non-Canonical Mechanisms

Recent research has revealed that splicing can be regulated by proteins with previously characterized functions in other processes. A study of the RNA editing enzyme ADAR1 found that the ADAR1p110 isoform influences alternative splicing independently of its canonical editing activity. The study found that ADAR1 interacts with pivotal spliceosome components and auxiliary splicing regulators and that ADAR1 extensively modulates alternative splicing events, with most alterations occurring independently of its RNA editing activity [<a href="#ref-12">12</a>].

Similarly, a study of histone demethylases KDM3A and KDM3B found that these proteins interact with RNA processing factors and that acute degradation of the endogenous proteins resulted in altered splicing independent of histone methylation status or catalytic activity. The splicing changes partially resembled the splicing pattern of the more blastocyst-like ground state of pluripotency [<a href="#ref-13">13</a>].

These findings have implications for researchers using DEXSeq or rMATS. If splicing can be regulated by proteins with non-canonical functions, then splicing changes detected by these tools may reflect complex regulatory mechanisms that are not immediately apparent from the known functions of the proteins under study.

Records and Documentation

Analysis Records

Maintain detailed records of the analysis workflow, including software versions, parameter settings, and annotation versions. The nf-core documentation emphasizes community pipeline standards for usage, configuration, and reproducible workflow context [<a href="#ref-7">7</a>]. Following these standards helps ensure that analyses can be reproduced by other researchers.

Data Management

Document the source of all input data, including the sequencing facility, library preparation method, and sequencing platform. The EMBL-EBI Training provides bioinformatics learning pathways and data resource training that can help researchers develop consistent data management practices [<a href="#ref-14">14</a>].

Reproducibility Practices

Use version control for analysis scripts and document all parameter choices. The Carpentries lessons provide foundational computing, data, shell, Git, and programming training that supports reproducible analysis practices [<a href="#ref-10">10</a>]. Researchers should also consider using workflow management systems to automate and document analysis steps.

Common Failure Patterns

Failure to Match Tool to Biological Question

A common failure is using DEXSeq when the biological question concerns specific splicing events or using rMATS when the question concerns broader exon usage changes. This mismatch can lead to results that do not directly address the research question or that miss important biological signals.

Inadequate Replication

Both DEXSeq and rMATS require biological replicates to estimate variability and perform statistical testing. Inadequate replication can lead to false positive or false negative results. Researchers should design experiments with sufficient replication based on the expected effect size and variability.

Poor Annotation Quality

Using outdated or incomplete gene annotations can lead to incorrect exon bin construction in DEXSeq or incorrect event definition in rMATS. Researchers should use the most current annotation for their organism and should verify that the annotation is appropriate for their experimental context.

Ignoring Batch Effects

Batch effects can confound differential exon usage or splicing analysis if they are correlated with the conditions of interest. Researchers should include batch information in the statistical model when using DEXSeq and should consider batch effects when interpreting rMATS results.

Overinterpreting Small Delta PSI Values

Small delta PSI values may be statistically significant with large sample sizes but may not be biologically meaningful. Researchers should consider the biological context and validate findings with independent methods before drawing conclusions.

Failure to Validate Findings

Published studies demonstrate the importance of validation. In the Fuchs endothelial corneal dystrophy study, qPCR validation revealed elevated FN1 expression in patient samples even though DEXSeq analysis showed reduced expression across FN1 gene regions [<a href="#ref-3">3</a>]. Without validation, this discrepancy might have led to incorrect conclusions.

Limitations and Interpretation Boundaries

Benchmarking Variability

The review of differential splicing tools notes that benchmarking studies show no consensus on tool performance, revealing considerable variability across different scenarios [<a href="#ref-1">1</a>]. This finding means that researchers cannot assume that one tool will perform best in their specific context. The review recommends tools with high citation frequency and continued developer maintenance, such as DEXSeq and rMATS [<a href="#ref-1">1</a>].

Short-Read Limitations

Short-read RNA sequencing has inherent limitations in resolving complex splicing patterns. The central challenges in quantifying alternative splicing include correct splice site identification and accurate isoform deconvolution of transcripts [<a href="#ref-1">1</a>]. Researchers should be aware that short-read data may not fully resolve complex isoform structures.

Complementary Use of Long-Read Sequencing

Long-read RNA sequencing technologies are discussed as a complement to short-read methods, promising reduced deconvolution needs and further innovation [<a href="#ref-1">1</a>]. Researchers with access to long-read sequencing should consider whether it might provide additional information beyond what short-read analysis can reveal.

Interpretation of Exon Usage Changes

DEXSeq detects changes in exon usage that can arise from multiple mechanisms, including alternative splicing, alternative promoter usage, and alternative polyadenylation. Researchers should not assume that all DEXSeq results reflect splicing changes. The Fuchs endothelial corneal dystrophy study demonstrated that DEXSeq results can be paradoxical when compared to validation data [<a href="#ref-3">3</a>].

Quality Controls and Professional Escalation

Quality Control Checks

Perform quality control at multiple stages of the analysis. Check alignment quality, verify sample identity, and examine the distribution of counts before running DEXSeq or rMATS. The Galaxy Training Network provides accessible workflow training and analysis tutorials that include quality control steps for RNA-seq data [<a href="#ref-9">9</a>].

When to Escalate to Professional Support

Consider escalating to professional bioinformatics support when the analysis requires advanced statistical modeling, when results are inconsistent across tools or validation methods, or when the biological interpretation requires specialized expertise. The EMBL-EBI Training provides bioinformatics learning pathways that can help researchers identify when they need additional training or support [<a href="#ref-14">14</a>].

Documentation of Escalation Decisions

Document the reasons for escalation and the outcomes of professional consultations. This documentation supports transparency and reproducibility in the research process.

Safety and Regulatory Context

Data Privacy and Security

RNA-seq data may contain sensitive information about research participants. Researchers should follow institutional guidelines for data storage, sharing, and publication. The NCBI provides official descriptions of data resources and search systems that can help researchers identify appropriate data repositories [<a href="#ref-8">8</a>].

Compliance with Funding Requirements

Many funding agencies require data sharing and reproducible analysis practices. Researchers should be aware of the specific requirements for their funding sources and should document their analysis workflows accordingly.

Ethical Use of Public Data

When using public data sets, researchers should follow the data use agreements and citation requirements specified by the data producers. The NCBI provides official descriptions of database search systems and sequence resources that can help researchers identify appropriate public data sets [<a href="#ref-8">8</a>].

A Practical Decision Framework for DEXSeq and rMATS Based on Experimental Design

Mapping Experimental Designs to Tool Selection

The choice between DEXSeq and rMATS becomes more tractable when you map your specific experimental design to the analytical capabilities of each tool. The review of 22 differential splicing tools notes that benchmarking studies show no consensus on tool performance, revealing considerable variability across different scenarios [<a href="#ref-1">1</a>]. This variability means that the experimental design itself should drive the tool selection instead of relying on general recommendations.

For experiments with two conditions and biological replicates, both tools can perform adequately. The deciding factor becomes the granularity of the biological question. If you need to know whether specific splicing events change, rMATS provides event-level output with percent spliced in values. If you need to know whether any exon within a gene changes in relative usage, DEXSeq provides exon-level output across the entire gene body.

For time course experiments, the choice becomes more nuanced. DEXSeq can model time as a continuous or categorical variable within its generalized linear model framework, allowing you to test for exons whose usage changes over time. rMATS can compare splicing events between time points, but the interpretation of multiple pairwise comparisons becomes cumbersome. The adipocyte differentiation study used a differentiation time course and found that splicing was highly dynamic across differentiation, with the dynamics impacted by metabolic phenotype [<a href="#ref-2">2</a>]. This study demonstrates that time course designs can reveal splicing dynamics that static comparisons miss, but the analytical approach must accommodate the temporal structure.

For experiments with multiple factors, such as genotype and treatment, DEXSeq offers greater flexibility because its generalized linear model can include interaction terms. This allows you to test whether the effect of treatment on exon usage differs between genotypes. rMATS can also handle multi-factor designs, but the statistical model becomes more complex and the output requires careful interpretation.

For experiments with small effect sizes, the choice of tool can influence detection power. The Hub1 overexpression study detected seven exon skipping events with rMATS, with DYN2 showing significant differential splicing at an FDR of 0.0481 and a delta PSI of negative 0.036 [<a href="#ref-4">4</a>]. This small delta PSI value demonstrates that rMATS can detect statistically significant changes even when the biological effect is modest. However, the biological relevance of such small changes requires careful consideration.

A Structured Decision Matrix for Tool Selection

A practical decision matrix can help researchers match their experimental design to the appropriate tool. The matrix considers four dimensions: biological question, experimental design, data availability, and interpretation requirements.

For the biological question dimension, ask whether you need to identify specific splicing event types or whether you need to detect any change in exon usage. If the question concerns specific events such as skipped exons or retained introns, rMATS is the appropriate choice. If the question concerns broader changes in exon usage that may arise from multiple mechanisms, DEXSeq is appropriate.

For the experimental design dimension, consider the number of conditions, the presence of covariates, and the structure of the data. DEXSeq accommodates complex designs with multiple factors and covariates within its generalized linear model framework. rMATS handles simpler designs more directly, though it can accommodate multi-group comparisons.

For the data availability dimension, consider whether you have exon-level count data or aligned reads in BAM format. DEXSeq requires exon count bins constructed from aligned reads. rMATS requires aligned reads and a GTF annotation file. If you already have exon count data from a previous analysis, DEXSeq may be more efficient. If you have aligned reads and need event-level output, rMATS is appropriate.

For the interpretation requirements dimension, consider whether you need percent spliced in values for mechanistic interpretation or whether exon-level statistics suffice. rMATS provides PSI values that directly quantify the proportion of transcripts including a particular exon or splice site. DEXSeq provides exon usage statistics that require additional interpretation to connect to splicing mechanisms.

Implementing the Decision Framework in Practice

The practical implementation of this decision framework begins with writing a clear statement of the biological question before any analysis. This statement should specify whether the question concerns specific splicing events, broad exon usage changes, or both. The statement should also specify the experimental design, including the number of conditions, replicates, and covariates.

After writing the biological question, assess the experimental design against the capabilities of each tool. For two-group comparisons with adequate replication, both tools can work. For complex designs with multiple factors and covariates, DEXSeq offers greater flexibility. For designs where specific splicing events are the primary interest, rMATS provides more direct output.

The next step is to prepare the input data according to the requirements of the chosen tool. For DEXSeq, construct exon counting bins using the DEXSeq package and the gene annotation. The Bioconductor documentation provides official package documentation, installation instructions, and reproducible genomic analysis workflows [<a href="#ref-6">6</a>]. For rMATS, prepare the GTF annotation file and verify that it contains the necessary information for event definition.

After running the analysis, interpret the results in the context of the biological question. For DEXSeq, examine which exons show differential usage and whether the changes cluster in specific gene regions. For rMATS, examine which event types show differential splicing and whether the delta PSI values are biologically meaningful.

A Record System for Tool Selection Decisions

Maintaining a structured record of tool selection decisions supports reproducibility and helps researchers justify their analytical choices. The record should document the biological question, the experimental design, the tool selected, the rationale for the selection, and the alternative tools considered.

The record should also document the specific parameters used for each tool, including the annotation version, the read counting method, and the statistical thresholds. The nf-core documentation emphasizes community pipeline standards for usage, configuration, and reproducible workflow context [<a href="#ref-7">7</a>]. Following these standards helps ensure that analyses can be reproduced by other researchers.

A practical record system includes a decision log that captures the following information for each analysis: the date of the decision, the researcher making the decision, the biological question, the experimental design, the tool selected, the rationale for the selection, the alternative tools considered, and the reasons for rejecting the alternatives. This log serves as a reference for future analyses and provides transparency for collaborators and reviewers.

The record system should also include a parameter log that documents the specific settings used for each tool. This log should include the software version, the annotation version, the read length, the library type, and any filtering thresholds applied. The EMBL-EBI Training provides bioinformatics learning pathways and data resource training that can help researchers develop consistent data management practices [<a href="#ref-14">14</a>].

Troubleshooting Common Tool Selection Errors

Several common errors arise when researchers select between DEXSeq and rMATS. The first error is selecting a tool based on familiarity instead of the biological question. Researchers who have used one tool in previous analyses may default to that tool even when the current question requires the other. The decision framework described above helps avoid this error by forcing explicit consideration of the biological question.

The second error is assuming that both tools will produce concordant results. The Fuchs endothelial corneal dystrophy study demonstrated that DEXSeq analysis showed reduced expression across FN1 gene regions in patient samples, while subsequent qPCR validation demonstrated significantly elevated FN1 expression [<a href="#ref-3">3</a>]. This paradoxical finding highlights the importance of understanding what each tool measures and how the results relate to biological reality.

The third error is using both tools without a clear strategy for integrating the results. Some studies use both tools to gain complementary information, but the integration requires careful thought. In the Fuchs endothelial corneal dystrophy study, rMATS identified 486 upregulated and 89 downregulated intron retention events, while DEXSeq analysis showed reduced expression across FN1 gene regions [<a href="#ref-3">3</a>]. The integration of these results required understanding that intron retention and exon usage measure different aspects of RNA processing.

The fourth error is failing to validate tool selection decisions with independent methods. The sheep fat-tail study used rMATS alongside Whippet and applied stringent filters to identify high-confidence differential splicing events [<a href="#ref-5">5</a>]. This multi-tool strategy increased confidence in the results but required additional analytical effort.

Comparing Output Interpretation Workflows

The interpretation workflow for DEXSeq output differs substantially from the workflow for rMATS output. DEXSeq output consists of exon-level statistics that indicate whether the relative usage of each exon changes between conditions. The interpretation requires examining which exons show significant changes and whether the changes cluster in specific gene regions. This clustering can indicate coordinated regulation of exon usage, which may reflect alternative promoter usage, alternative polyadenylation, or splicing changes.

The Fuchs endothelial corneal dystrophy study provides an example of DEXSeq interpretation. DEXSeq analysis showed reduced expression across FN1 gene regions in patient samples [<a href="#ref-3">3</a>]. This pattern of reduced expression across multiple exons suggested a coordinated change in FN1 processing, though the subsequent validation revealed a more complex picture.

The rMATS interpretation workflow focuses on event-level statistics with percent spliced in values. The output identifies specific splicing events and provides delta PSI values that quantify the change in inclusion between conditions. The interpretation requires examining which event types show significant changes and whether the delta PSI values are biologically meaningful.

The Hub1 overexpression study provides an example of rMATS interpretation. rMATS detected seven exon skipping events, with DYN2 showing significant differential splicing at an FDR of 0.0481 and a delta PSI of negative 0.036 [<a href="#ref-4">4</a>]. MaxEntScan analysis confirmed that DYN2's 5' splice site was significantly weaker than canonical yeast splice sites, connecting the splicing change to the underlying sequence features [<a href="#ref-4">4</a>].

Integrating Tool Selection with Downstream Analyses

The choice between DEXSeq and rMATS affects downstream analyses, including functional enrichment, network analysis, and variant interpretation. The sheep fat-tail study used rMATS to identify differential splicing events and then performed functional enrichment analysis, which highlighted the importance of the genes in the underlying molecular mechanisms of fat metabolism [<a href="#ref-5">5</a>]. The study also performed protein-protein interaction network analysis, which revealed that the differentially spliced genes were significantly connected [<a href="#ref-5">5</a>].

The adipocyte differentiation study integrated splicing results with genome-wide association study data and found that type 2 diabetes associated variants were enriched in regions that were differentially spliced in early differentiation [<a href="#ref-2">2</a>]. This integration required splicing results that could be mapped to genomic regions, which is more straightforward with rMATS event-level output than with DEXSeq exon-level output.

The breast cancer study used alternative splicing analysis to identify splicing subtypes in older adults patients and then used machine learning to construct a predictive model [<a href="#ref-11">11</a>]. This analysis required splicing results that could be used as features in the predictive model, which is more direct with rMATS event-level output.

A Comparison of Statistical Power Considerations

Statistical power differs between DEXSeq and rMATS due to their different analytical approaches. DEXSeq tests each exon within a gene, which involves multiple testing across all exons in the genome. rMATS tests each splicing event, which also involves multiple testing across all detected events. The multiple testing burden differs between the tools because the number of tests differs.

The review of differential splicing tools notes that benchmarking studies show no consensus on tool performance, revealing considerable variability across different scenarios [<a href="#ref-1">1</a>]. This variability means that power calculations for one tool may not transfer to the other. Researchers should consider the specific characteristics of their data when estimating power.

For experiments with limited replication, the choice of tool can affect the ability to detect significant changes. Both tools require biological replicates to estimate variability and perform statistical testing. The number of replicates needed depends on the expected effect size and variability in the experimental system.

A Practical Checklist for Tool Selection

A practical checklist can help researchers work through the tool selection decision systematically. The checklist begins with the biological question and proceeds through the experimental design, data availability, and interpretation requirements.

First, write the biological question in a single sentence. Specify whether the question concerns specific splicing events, broad exon usage changes, or both. If the question concerns specific events, rMATS is the appropriate choice. If the question concerns broad exon usage changes, DEXSeq is appropriate.

Second, describe the experimental design, including the number of conditions, replicates, and covariates. If the design includes multiple factors or covariates, DEXSeq offers greater flexibility. If the design is a simple two-group comparison, both tools can work.

Third, assess the available data. If you have exon-level count data, DEXSeq may be more efficient. If you have aligned reads in BAM format, both tools can work. Verify that the annotation file is current and appropriate for your organism.

Fourth, consider the interpretation requirements. If you need percent spliced in values for mechanistic interpretation, rMATS provides these directly. If exon-level statistics suffice, DEXSeq is appropriate.

Fifth, document the decision in the record system, including the rationale and the alternatives considered. This documentation supports reproducibility and provides transparency for collaborators and reviewers.

Using the Decision Framework with Published Examples

The decision framework can be applied to published studies to illustrate how tool selection follows from the biological question and experimental design. The Hub1 overexpression study used rMATS to detect exon skipping events in Saccharomyces cerevisiae [<a href="#ref-4">4</a>]. The biological question concerned specific splicing events, and the experimental design compared Hub1 overexpression to control conditions. The use of rMATS aligned with the need for event-level output with PSI values.

The sheep fat-tail study used rMATS alongside Whippet to identify differential splicing events between fat-tailed and thin-tailed sheep breeds [<a href="#ref-5">5</a>]. The biological question concerned specific splicing events involved in fat-tail development, and the experimental design compared two breed types across multiple datasets. The use of rMATS aligned with the need for event-level output that could be filtered for high-confidence events.

The Fuchs endothelial corneal dystrophy study used both rMATS and DEXSeq to investigate splicing changes [<a href="#ref-3">3</a>]. The biological question concerned both specific splicing events and broader exon usage changes. The use of both tools provided complementary information, with rMATS identifying intron retention events and DEXSeq showing reduced expression across FN1 gene regions.

The adipocyte differentiation study used splicing analysis to characterize splicing dynamics across differentiation [<a href="#ref-2">2</a>]. The biological question concerned the dynamics of splicing across a time course, and the experimental design included multiple time points and metabolic phenotypes. The study found little overlap between genes that were differentially spliced and those that were differentially expressed, positioning alternative splicing as an independent regulatory mechanism [<a href="#ref-2">2</a>].

A Troubleshooting Method for Discordant Results

When DEXSeq and rMATS produce discordant results, a structured troubleshooting method can help resolve the discrepancy. The first step is to verify that both tools used the same annotation version and the same aligned reads. Differences in annotation or alignment can produce discordant results.

The second step is to examine the specific genomic regions where the results differ. For the Fuchs endothelial corneal dystrophy study, DEXSeq analysis showed reduced expression across FN1 gene regions, while qPCR validation demonstrated elevated FN1 expression [<a href="#ref-3">3</a>]. This discrepancy was resolved by understanding that the loss of normal intron retention mediated regulation may contribute to pathological FN1 overexpression [<a href="#ref-3">3</a>].

The third step is to consider whether the tools are measuring different biological phenomena. DEXSeq measures exon usage, which can change due to alternative splicing, alternative promoter usage, or alternative polyadenylation. rMATS measures specific splicing events. Discordant results may indicate that the biological change involves a mechanism that one tool captures and the other does not.

The fourth step is to validate the discordant results with an independent method. Quantitative PCR can confirm changes in isoform expression or exon inclusion. In the Fuchs endothelial corneal dystrophy study, qPCR validation in an independent cohort demonstrated elevated FN1 expression even though DEXSeq analysis showed reduced expression across FN1 gene regions [<a href="#ref-3">3</a>].

A Comparison of Filtering Strategies

Filtering strategies differ between DEXSeq and rMATS analyses. For DEXSeq, filtering typically involves removing exons with low counts and applying significance thresholds to the exon usage statistics. For rMATS, filtering typically involves removing events with low read support and applying significance thresholds to the splicing statistics.

The sheep fat-tail study applied stringent filters to the rMATS results, only considering events detected by both rMATS and Whippet in at least three datasets with the same delta PSI trend [<a href="#ref-5">5</a>]. This filtering strategy increased confidence in the detected events but reduced the number of events considered.

The Fuchs endothelial corneal dystrophy study used a more stringent algorithm, IRFinder, to validate the intron retention events identified by rMATS [<a href="#ref-3">3</a>]. This validation step reduced the number of events from 486 upregulated and 89 downregulated to 10 upregulated events across nine genes and 6 downregulated events exclusively localized within FN1 [<a href="#ref-3">3</a>].

A Practical Approach to Reporting Tool Selection

Reporting tool selection in publications requires transparency about the decision process and the parameters used. The methods section should state the biological question, the tool selected, the rationale for the selection, and the specific parameters used. The nf-core documentation emphasizes community pipeline standards for usage, configuration, and reproducible workflow context [<a href="#ref-7">7</a>].

The methods section should also document the annotation version, the read counting method, and the statistical thresholds. This documentation allows other researchers to reproduce the analysis and to understand the basis for the conclusions.

The limitations section should acknowledge the constraints of the chosen tool and the potential for discordant results with other tools. The review of differential splicing tools notes that benchmarking studies show no consensus on tool performance, revealing considerable variability across different scenarios [<a href="#ref-1">1</a>]. Acknowledging this variability supports honest reporting of analytical choices.

Frequently Asked Questions

What is the main difference between DEXSeq and rMATS?

DEXSeq tests for differential exon usage within genes, detecting changes in the relative abundance of exons between conditions. rMATS detects specific splicing event types, including skipped exons, alternative splice sites, mutually exclusive exons, and retained introns, and quantifies the percent spliced in value for each event. The choice depends on whether the biological question concerns broad exon usage changes or specific splicing events.

Can I use both DEXSeq and rMATS in the same study?

Yes, several published studies have used both tools to gain complementary information. In a study of Fuchs endothelial corneal dystrophy, rMATS identified intron retention events while DEXSeq showed reduced expression across FN1 gene regions, providing a more complete picture than either tool alone [<a href="#ref-3">3</a>]. Using both tools can help researchers understand both the specific splicing events and the broader exon usage changes.

What input data do I need for DEXSeq?

DEXSeq requires exon-level count data generated from aligned RNA-seq reads in BAM format. The DEXSeq package constructs exon counting bins from a gene annotation file and counts reads overlapping each bin. The Bioconductor documentation provides official package documentation and reproducible genomic analysis workflows [<a href="#ref-6">6</a>].

What input data do I need for rMATS?

rMATS requires aligned RNA-seq reads in BAM format along with a gene annotation file in GTF format. The tool uses the annotation to define splicing events and counts reads that support inclusion or exclusion of each event. rMATS can accept single-end or paired-end reads and requires information about read length and library type.

How do I interpret delta PSI values from rMATS?

Delta PSI represents the change in the percent spliced in value between conditions. A negative delta PSI indicates decreased exon inclusion, while a positive delta PSI indicates increased exon inclusion. In the Hub1 overexpression study, DYN2 showed a delta PSI of negative 0.036, indicating a small decrease in exon inclusion [<a href="#ref-4">4</a>]. Researchers should consider whether the magnitude of delta PSI is biologically meaningful in their context.

What are the limitations of short-read RNA sequencing for splicing analysis?

Short-read RNA sequencing has limitations in resolving complex splicing patterns. The central challenges include correct splice site identification and accurate isoform deconvolution of transcripts [<a href="#ref-1">1</a>]. Long-read sequencing technologies are emerging as a complement to short-read methods, promising reduced deconvolution needs [<a href="#ref-1">1</a>].

How many biological replicates do I need for DEXSeq or rMATS?

The number of biological replicates depends on the expected effect size and variability in the experimental system. Both tools require replicates to estimate variability and perform statistical testing. Researchers should design experiments with sufficient replication based on their specific experimental context and should consult the tool documentation for guidance.

How do I validate DEXSeq or rMATS results?

Validation can be performed using independent methods such as quantitative PCR to confirm changes in isoform expression or exon inclusion. In the Fuchs endothelial corneal dystrophy study, qPCR validation in an independent cohort demonstrated elevated FN1 expression even though DEXSeq analysis showed reduced expression across FN1 gene regions [<a href="#ref-3">3</a>]. This example highlights the importance of validation for interpreting results correctly.

Related Bioinformatics Guides

• RNA-Seq Alignment: Choosing the Right Tool and Parameters
• Gene Set Enrichment Analysis Tools: Choosing the Right One
• Multi-Omics Data Integration: A Comparative Framework for Choosing the Right Method
• Metagenomics vs Metabarcoding: Choosing the Right Approach for Your Study
• Genomic Data Analysis Tools: A Comparative Guide for Researchers

Related Clinical & Scientific Guides

• A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
• Computational Immunology: Modeling the Immune System
• How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices

References and Further Reading

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.