Allele-Specific Expression Analysis with Long-Read RNA-Seq: A Guide to Phased Transcriptomes

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-21

Allele-specific expression (ASE) analysis with long-read RNA sequencing requires a phased genome reference, haplotype-aware read alignment, and isoform-level quantification to assign transcripts to parental chromosomes. This guide provides a practical workflow for researchers who need to move from raw long-read data to biologically interpretable ASE results, covering genome phasing with Whatshap, read alignment, and quantification with Salmon, along with quality controls, common failure patterns, and interpretation limits.

Key Takeaways

• Allele-specific expression analysis with long-read RNA-seq requires a phased genome reference, haplotype-aware read alignment, and isoform-level quantification to assign transcripts to parental chromosomes.
• Whatshap phases genomic variants into haplotypes using DNA sequencing reads, and phase blocks that cover complete genes are critical for confident read assignment in ASE quantification.
• Long reads spanning entire cDNA molecules can cover multiple heterozygous variants, enabling confident haplotype assignment and detection of both allele-specific expression and allele-specific transcript structure.
• The workflow includes preparing a reference genome, phased VCF, and long-read RNA-seq data, then aligning reads to two haplotype-specific references and assigning each read to the better-matching haplotype.
• Salmon quantifies transcript counts separately for each haplotype, and statistical testing for allelic imbalance compares haplotype A and B counts, with tools like LORALS and LongPolyASE designed for long-read data.

Scope and Reader Context

This article addresses the specific problem of quantifying allele-specific expression using long-read RNA-seq data. The workflow assumes you have access to either Oxford Nanopore Technologies (ONT) or Pacific Biosciences (PacBio) sequencing output, a reference genome, and a variant call set from the same individual or population. The target reader is a graduate student, postdoctoral researcher, laboratory professional, or bioinformatics practitioner who understands basic RNA-seq analysis but has not yet implemented haplotype-aware ASE quantification.

The practical outcome is a reproducible protocol that produces phased transcriptomes, assigns long reads to haplotypes, and generates allele-specific counts at both gene and isoform levels. The workflow integrates established tools including Whatshap for variant phasing, alignment tools compatible with long reads, and Salmon for transcript-level quantification. The guide also covers quality checks, common failure modes, and when to escalate technical problems to specialized support.

Why Long Reads Change Allele-Specific Expression Analysis

Short-read RNA-seq has been the standard for gene expression quantification for years, but it has a fundamental limitation for ASE analysis. Short reads typically span only a fraction of a transcript, so assigning a read to a specific haplotype requires that the read overlaps a heterozygous variant. When variants are sparse or when the read does not cover the variant position, the read cannot be assigned to either parental chromosome. This leads to unassigned reads and reduced statistical power for detecting allelic imbalance.

Long-read sequencing platforms from Oxford Nanopore Technologies and Pacific Biosciences produce reads that span entire cDNA molecules. This property allows a single read to cover multiple heterozygous variants along the transcript, enabling confident haplotype assignment. The resolution gained from long reads has been demonstrated in large human datasets, where allele-specific analysis of long reads revealed transcript alterations caused by both common and rare genetic variants that would be difficult or impossible to detect with short reads alone [<a href="#ref-1">1</a>]. The same study identified over 70,000 novel transcripts for annotated genes, underscoring the isoform-level information that long reads provide beyond simple gene expression counts.

The practical consequence for ASE analysis is that long reads enable two distinct types of allele-specific information. First, allele-specific expression quantifies whether one haplotype produces more total transcript than the other. Second, allele-specific transcript structure identifies cases where the two haplotypes produce different isoforms from the same gene, such as one haplotype including a particular exon while the other excludes it. Both types of information are biologically meaningful and can reveal cis-regulatory effects, splicing differences, and disease-associated variant effects [<a href="#ref-2">2</a>].

Core Principles of Phased Transcriptome Analysis

Haplotype Phasing as the Foundation

Before any RNA-seq read can be assigned to a parental chromosome, the genomic variants must be organized into haplotypes. Phasing is the process of determining which variants are inherited together on the same chromosome copy. For ASE analysis, phasing provides the reference against which RNA-seq reads are compared.

Whatshap is a widely used tool for haplotype phasing that works with both short-read and long-read sequencing data. The tool uses read information to connect variants into phase blocks, which are contiguous genomic regions where the phase is known. For RNA-seq ASE analysis, the phase information from genomic DNA sequencing is typically used, because RNA-seq reads alone may not provide complete phasing coverage across all expressed genes.

The quality of phasing directly affects ASE quantification. If variants are incorrectly phased, reads will be assigned to the wrong haplotype, producing false allelic imbalance or masking true imbalance. Researchers should assess phase block length and switch error rates before proceeding to ASE quantification. Phase blocks that cover entire genes are ideal, while fragmented phase blocks that split genes reduce the ability to assign reads across the full transcript length.

Reference Genome Requirements

A haplotype-resolved reference genome is the ideal substrate for ASE analysis. Haplotype-resolved assemblies provide two complete chromosome copies, allowing reads to be aligned directly to each haplotype. This approach eliminates the need to infer haplotypes from variant calls and provides the most direct assignment of reads to parental chromosomes.

When a haplotype-resolved reference is not available, researchers can construct haplotype-specific references by substituting the phased variants into the linear reference genome. This produces two reference sequences, one for each haplotype, and reads are aligned to both. The alignment results are then compared to assign each read to the haplotype with the better match.

The choice between these approaches depends on data availability. Haplotype-resolved assemblies are becoming more common but are still not available for every species or individual. The variant-substitution approach is more widely applicable because it only requires a standard reference genome and a phased variant call set.

Isoform-Level Resolution

ASE analysis at the isoform level adds a layer of complexity beyond gene-level quantification. A gene may produce multiple isoforms, and the allelic balance may differ between isoforms. For example, one haplotype may produce predominantly the full-length isoform while the other haplotype produces predominantly a truncated isoform. Gene-level ASE analysis would miss this pattern because the total expression from each haplotype could be balanced.

Long-read sequencing is particularly suited to isoform-level ASE because full-length reads capture complete transcript structures. The LORALS computational package, developed for allele-specific analysis of long reads, characterizes both allele-specific expression and allele-specific transcript structure events [<a href="#ref-1">1</a>]. This dual capability is a key advantage of long-read approaches over short-read methods.

For polyploid organisms, isoform-level ASE analysis requires additional considerations. The LongPolyASE framework addresses this gap by enabling allele-specific gene and isoform expression analysis in diploid and polyploid organisms using long-read RNA-seq [<a href="#ref-3">3</a>]. The framework includes components for identifying syntenic gene relationships, discovering novel isoforms, and performing statistical testing of allelic imbalance and isoform usage. Its application to rice, potato, rapeseed, and strawberry demonstrates that the approach works across ploidy levels and sequencing platforms.

At a Glance: Workflow Components and Decision Points

Practical Workflow for Phasing and ASE Quantification

Step 1: Prepare Input Data

The workflow requires three primary inputs. First, a reference genome in FASTA format. Second, a variant call set in VCF format from the same individual or a closely related individual. Third, long-read RNA-seq data in FASTA or FASTQ format from the same individual.

The variant call set should be phased before proceeding. If the variant call set is not phased, run Whatshap using genomic DNA sequencing data from the same individual. The phasing step requires read data that spans multiple variants, so whole-genome sequencing data is preferred over exome or targeted sequencing data.

For RNA-seq data, the sequencing platform choice affects downstream analysis. PacBio data generally provides higher base-level accuracy than ONT data, which translates to more accurate novel transcript identification and allele-specific gene and isoform expression calls [<a href="#ref-4">4</a>]. However, ONT data is also usable, and the choice between platforms often depends on cost, throughput, and instrument availability.

Step 2: Phase the Genome

Run Whatshap on the variant call set using the genomic DNA sequencing reads. The output is a phased VCF file where each variant is assigned to a haplotype. Assess the quality of phasing by examining phase block lengths and the number of variants per block.

For ASE analysis, the critical quality metric is whether phase blocks cover complete genes of interest. If a gene is split across multiple phase blocks, reads from that gene cannot be assigned to haplotypes with confidence across the full transcript length. In this case, consider whether additional sequencing depth or a different phasing approach would improve phase block continuity.

Step 3: Construct Haplotype-Specific References

If a haplotype-resolved reference genome is available, use it directly. The two haplotypes serve as the reference sequences for read alignment.

If only a linear reference genome is available, construct two haplotype-specific references by substituting the phased variants into the reference. This produces reference sequence A and reference sequence B, each containing the variants assigned to that haplotype. Tools for this substitution are available in standard bioinformatics toolkits, and the process is straightforward for single nucleotide variants and small indels.

For structural variants, the substitution approach is more complex. Large deletions, insertions, duplications, and inversions cannot be represented by simple variant substitution. In these cases, a haplotype-resolved assembly is strongly preferred, or the structural variant regions should be excluded from ASE analysis.

Step 4: Align RNA-Seq Reads

Align the long-read RNA-seq data to both haplotype-specific references. The alignment should be performed separately for each reference, producing two alignment files. Use a long-read aligner that is appropriate for the sequencing platform. The aligner should be configured for spliced alignment because RNA-seq reads span exon junctions.

After alignment, compare the mapping quality and alignment scores for each read against the two references. Assign each read to the haplotype with the better alignment. Reads that align equally well to both haplotypes or that do not meet a minimum quality threshold should be excluded from ASE quantification.

The mapping rate is an important quality metric. Low mapping rates may indicate contamination, adapter issues, or reference problems. High mapping rates with poor haplotype assignment may indicate that the sample has low heterozygosity or that the phasing is incorrect.

Step 5: Quantify Transcripts Per Haplotype

With reads assigned to haplotypes, quantify transcript expression using Salmon or an equivalent transcript quantification tool. The quantification should be performed separately for each haplotype, producing transcript-level counts for haplotype A and haplotype B.

Salmon requires a transcript index that is built from the reference transcript sequences. For haplotype-aware quantification, build separate indexes for each haplotype or use a combined index that distinguishes transcripts by haplotype origin.

The output is a count matrix with transcripts as rows and samples as columns, where each sample is a haplotype. For a single individual, this produces two columns representing the two haplotypes. For multiple individuals, the matrix expands accordingly.

Step 6: Test for Allelic Imbalance

The final step is statistical testing for allelic imbalance. For each gene or transcript, compare the counts from haplotype A and haplotype B. The appropriate statistical test depends on the experimental design and the number of biological replicates.

For a single individual, a simple test of proportion can identify genes where the allelic ratio deviates significantly from 1:1. For multiple individuals, mixed models or other approaches that account for between-individual variation are more appropriate.

The LORALS package provides allele-specific analysis specifically designed for long-read data [<a href="#ref-1">1</a>]. The LongPolyASE framework provides statistical testing and visualization of allelic imbalance and isoform usage for diploid and polyploid organisms [<a href="#ref-3">3</a>]. These tools handle the specific challenges of long-read data, including variable read lengths and sequencing errors.

Options and Tradeoffs in Workflow Design

Platform Choice: ONT versus PacBio

The choice between ONT and PacBio sequencing affects data quality and downstream analysis. A systematic evaluation of both platforms using the same single-cell cDNA libraries found that PacBio outperformed ONT in the accuracy of novel transcript identification and allele-specific gene and isoform expression [<a href="#ref-4">4</a>]. This difference is attributed to the higher sequencing quality of PacBio.

However, ONT offers advantages in throughput and cost for some applications. The choice should be guided by the specific research question. If the goal is to identify novel isoforms with high confidence, PacBio may be worth the additional cost. If the goal is to survey allele-specific expression across many samples, ONT may provide sufficient accuracy at lower cost.

Gene-Level versus Isoform-Level Analysis

Gene-level ASE analysis is simpler and requires less sequencing depth. Isoform-level analysis provides more biological detail but requires higher depth to achieve statistical power for each isoform.

The choice depends on the biological question. If the interest is in cis-regulatory effects on total gene expression, gene-level analysis is sufficient. If the interest is in splicing regulation or isoform switching, isoform-level analysis is necessary.

Long-read data is particularly valuable for isoform-level analysis because full-length reads capture complete transcript structures. Short-read data cannot reliably distinguish between isoforms that share exons, limiting the resolution of isoform-level ASE analysis.

Diploid versus Polyploid Analysis

Most ASE tools are designed for diploid organisms. Polyploid organisms present additional challenges because more than two haplotypes must be distinguished, and homeologous genes may share high sequence similarity.

The LongPolyASE framework addresses this gap by providing an end-to-end computational framework for allele-specific gene and isoform expression analysis in diploid and polyploid organisms [<a href="#ref-3">3</a>]. The framework includes components for identifying syntenic gene relationships, discovering novel isoforms, and performing statistical testing. Its successful application to autotetraploid potato, allotetraploid rapeseed, and allooctoploid strawberry demonstrates that the approach scales to complex genomes.

For researchers working with polyploid species, using a purpose-built framework is strongly recommended over adapting diploid tools. The additional complexity of polyploid analysis requires specialized methods for distinguishing homeologous transcripts and assigning reads to the correct subgenome.

Observations and Measurements for Quality Control

Sequencing Depth Requirements

The sequencing depth required for ASE analysis depends on the number of genes or transcripts of interest and the magnitude of allelic imbalance to be detected. Detecting small imbalances requires higher depth than detecting large imbalances.

For gene-level ASE analysis, a general guideline is to aim for sufficient depth to obtain at least 20 to 30 reads per gene per haplotype for genes of interest. This depth provides reasonable power to detect moderate allelic imbalance. For isoform-level analysis, higher depth is needed because reads are distributed across multiple isoforms.

Researchers should perform a power analysis before sequencing to determine the required depth. The power analysis should account for the expected read assignment rate, the number of genes of interest, and the minimum effect size to be detected.

Read Assignment Rate

The read assignment rate is the proportion of reads that can be confidently assigned to a haplotype. This rate depends on the density of heterozygous variants, the read length, and the sequencing error rate.

Long reads have higher assignment rates than short reads because they are more likely to span multiple heterozygous variants. However, assignment rates vary across genes depending on local variant density. Genes in regions with low heterozygosity will have lower assignment rates.

Monitor the assignment rate per gene and flag genes with low assignment rates. These genes should be excluded from ASE analysis or interpreted with caution because the counts may not accurately reflect the true allelic balance.

Mapping Quality Metrics

Standard alignment metrics apply to long-read RNA-seq data. These include the overall mapping rate, the proportion of reads mapped to each haplotype, and the distribution of mapping qualities.

A balanced mapping to both haplotypes is expected for most genes. Large deviations from balance across many genes may indicate a technical artifact, such as reference bias or alignment bias. Reference bias occurs when reads from one haplotype align preferentially to the reference because of sequence similarity. This bias can be reduced by using a haplotype-resolved reference or by using an alignment approach that is less sensitive to reference choice.

Novel Isoform Detection Rate

Long-read RNA-seq data often reveals novel isoforms that are not present in reference annotations. The rate of novel isoform detection is a useful quality metric because it confirms that the data contains full-length transcripts.

In a large human dataset, long-read RNA-seq identified over 70,000 novel transcripts for annotated genes, with protein expression validated for 10% of the novel transcripts [<a href="#ref-1">1</a>]. The detection of novel isoforms is expected in most long-read datasets and indicates that the sequencing captured complete transcripts.

For ASE analysis, novel isoforms should be included in the quantification because they may show allele-specific expression patterns that differ from annotated isoforms. Excluding novel isoforms could miss biologically important ASE events.

Records and Documentation for Reproducibility

Version Control for Tools and References

Reproducibility in ASE analysis requires careful documentation of all software versions, reference genome versions, and parameter settings. Small differences in tool versions can produce different results, so the exact versions must be recorded.

Use version control for analysis scripts and configuration files. The Carpentries lessons provide foundational training in version control with Git, which is essential for tracking changes to analysis code [<a href="#ref-5">5</a>]. The lessons also cover shell and programming fundamentals that are useful for implementing reproducible workflows.

Workflow Management Systems

Workflow management systems provide structure for complex analyses and ensure that each step is executed with the correct inputs and parameters. The nf-core documentation describes community standards for pipeline usage, configuration, and reproducible workflow context [<a href="#ref-6">6</a>]. These standards are useful for implementing ASE analysis pipelines that can be shared and reused.

The Galaxy Training Network (GTN) hosts peer-reviewed, browser-accessible tutorials that walk researchers through reproducible analysis pipelines.

Containerization

Containerization ensures that the analysis environment is identical across runs and across researchers. Tools such as Docker and Singularity package the software and its dependencies into a single image that can be run on any system.

For ASE analysis, containerization is particularly important because the workflow involves multiple tools with complex dependencies. A container that includes Whatshap, the aligner, Salmon, and the ASE testing tool ensures that all components are compatible and that the analysis can be reproduced.

Analysis Logs

Maintain a log of all analysis steps, including the commands executed, the input files, the output files, and any warnings or errors. The log should be detailed enough that another researcher could reproduce the analysis from the log alone.

The log should also record any deviations from the standard workflow. For example, if a particular gene was excluded because of low read assignment, this should be documented. If parameters were adjusted for a particular sample, this should be recorded.

Common Failure Patterns and Troubleshooting

Low Read Assignment to Haplotypes

Low read assignment rates can result from low heterozygosity, short phase blocks, or high sequencing error rates. If many reads cannot be assigned to a haplotype, the ASE analysis will have reduced power and may miss true allelic imbalance.

Troubleshooting steps include checking the heterozygosity of the sample, examining phase block lengths for genes of interest, and verifying that the phasing was performed correctly. If the sample has low heterozygosity, consider whether the research question can be addressed with the available data or whether additional samples with higher heterozygosity are needed.

Reference Bias

Reference bias occurs when reads from one haplotype align preferentially to the reference genome because of sequence similarity. This bias can produce false allelic imbalance or mask true imbalance.

Reference bias is reduced by using a haplotype-resolved reference genome. If a haplotype-resolved reference is not available, using a variant-substituted reference can reduce bias compared to aligning to the linear reference alone. The alignment approach should be tested for bias by examining the allelic balance in genes known to have balanced expression.

Phase Block Fragmentation

Fragmented phase blocks that split genes prevent confident haplotype assignment across the full transcript length. This problem is more common in genomic regions with low variant density or in samples with low sequencing depth for phasing.

Options for addressing phase block fragmentation include increasing the sequencing depth for phasing, using a different phasing tool, or using a haplotype-resolved reference genome. If the phase blocks cannot be improved, genes that are split across phase blocks should be excluded from ASE analysis.

Isoform Annotation Errors

Errors in isoform annotation can lead to incorrect quantification and false ASE calls. These errors are more common for novel isoforms that are not present in reference annotations.

The LongPolyASE framework includes a component for identifying annotation inconsistencies as part of its workflow [<a href="#ref-3">3</a>]. This component helps ensure that the isoform annotations used for quantification are accurate. Researchers should review novel isoform calls and validate a subset with orthogonal methods when possible.

Batch Effects in Multi-Sample Studies

When ASE analysis is performed across multiple samples, batch effects can introduce systematic biases. These effects can arise from differences in library preparation, sequencing runs, or sample processing.

Batch effects should be assessed and accounted for in the statistical analysis. Including batch as a covariate in the model can reduce the impact of batch effects. Alternatively, samples can be randomized across batches to balance the design.

Limitations of Long-Read ASE Analysis

Sequencing Error Rates

Long-read sequencing has higher error rates than short-read sequencing, particularly for ONT data. These errors can affect variant calling, read assignment, and isoform quantification.

The impact of sequencing errors on ASE analysis depends on the error rate and the analysis approach. Higher quality PacBio data generally provides more accurate results than ONT data for novel transcript identification and allele-specific expression [<a href="#ref-4">4</a>]. Researchers should be aware of the error profile of their chosen platform and adjust quality filters accordingly.

Coverage and Cost

Long-read sequencing is generally more expensive per base than short-read sequencing. Achieving sufficient coverage for ASE analysis across many genes or samples can be costly.

The cost consideration should be balanced against the biological value of the information gained. For some research questions, short-read sequencing may be sufficient, and the additional cost of long-read sequencing may not be justified. For questions that require isoform-level resolution or haplotype assignment across long distances, long-read sequencing is necessary.

Computational Requirements

Long-read RNA-seq data are large, and the analysis requires substantial computational resources. Alignment, quantification, and statistical testing can be computationally intensive, particularly for polyploid organisms with large genomes.

Researchers should plan for the computational requirements before starting the analysis. Cloud computing or institutional high-performance computing resources may be needed for large datasets. The Bioconductor project provides documentation for reproducible genomic analysis that includes guidance on managing computational resources [<a href="#ref-7">7</a>].

Interpretation Challenges

ASE analysis identifies allelic imbalance but does not directly reveal the mechanism. Imbalance can result from cis-regulatory variants, epigenetic differences, or post-transcriptional regulation. Distinguishing between these mechanisms requires additional experiments.

The biological interpretation of ASE results should be cautious. A finding of allelic imbalance is a starting point for further investigation, not a definitive conclusion about the underlying mechanism. Validation experiments, such as reporter assays or allele-specific chromatin analysis, are often needed to confirm the mechanism.

Safety and Regulatory Context

Data Privacy and Consent

RNA-seq data from human subjects contains genetic information that can identify individuals. Researchers must ensure that data collection and analysis comply with applicable privacy regulations and that participants have provided informed consent for the research use of their data.

NCBI's data portals provide step-by-step submission guides, data access policies, and API documentation for each of the institute's curated biological databases.

Data Sharing and Repositories

Sharing ASE analysis results and raw data supports reproducibility and enables meta-analyses. The NCBI provides databases for sequence data and analysis results [<a href="#ref-8">8</a>]. The EMBL-EBI provides training and data resources for bioinformatics analysis [<a href="#ref-9">9</a>].

Before sharing data, researchers should verify that the data can be shared under the consent agreements and applicable regulations. De-identification may be required for human data, and controlled access may be necessary for sensitive data.

Ethical Use of Findings

ASE analysis can reveal genetic variants associated with disease or other traits. Researchers should consider the ethical implications of their findings and avoid overinterpreting results that have not been validated.

Findings from ASE analysis should be reported with appropriate caveats about the limitations of the analysis. The clinical relevance of ASE findings should not be overstated without functional validation and replication in independent cohorts.

Professional Escalation Criteria

When to Seek Specialized Support

Some technical problems in ASE analysis require specialized expertise. Researchers should escalate to bioinformatics core facilities, tool developers, or specialized consultants when they encounter problems that they cannot resolve with standard troubleshooting.

Specific situations that warrant escalation include persistent low read assignment rates that do not improve with troubleshooting, phase block fragmentation that cannot be resolved with available data, and unexpected patterns of allelic imbalance across many genes that suggest a systematic technical artifact.

When to Consult Statistical Experts

The statistical analysis of ASE data can be complex, particularly for multi-sample studies or polyploid organisms. Researchers should consult statistical experts when the experimental design requires advanced statistical methods or when the results are difficult to interpret.

Statistical consultation is particularly important when the ASE analysis will be used for high-impact publications or for decisions with clinical or agricultural implications. A statistician can help ensure that the analysis is appropriate for the experimental design and that the conclusions are supported by the data.

When to Validate with Orthogonal Methods

ASE findings should be validated with orthogonal methods before drawing strong conclusions. Validation approaches include quantitative PCR with allele-specific primers, targeted sequencing with different technology, or functional assays.

The need for validation depends on the significance of the finding and the potential impact of a false positive. Findings that will be used for clinical decisions, breeding decisions, or high-impact publications should be validated. Findings that are exploratory and will be followed up in subsequent studies may not require immediate validation.

Decision Framework for Choosing Between Gene-Level and Isoform-Level ASE Analysis

The choice between gene-level and isoform-level allele-specific expression analysis is not a single binary decision but a series of decisions that depend on the biological question, the available sequencing depth, and the genomic complexity of the study system. This section provides a structured decision framework that researchers can apply before committing to a particular analysis strategy.

Decision Point 1: Define the Primary Biological Question

The first decision point requires a precise statement of the biological question. If the question concerns cis-regulatory effects on total transcript output from each haplotype, gene-level analysis is sufficient. If the question concerns splicing regulation, isoform switching, or differential isoform usage between haplotypes, isoform-level analysis is required.

A study of colorectal cancer used high-depth long-read single-cell RNA-seq to reveal extensive isoform diversities and splicing alterations, including widespread 3-prime UTR shortening and reduced intron retention [<a href="#ref-10">10</a>]. The same study performed allele-specific expression analysis that revealed dominant mutant allele expression in key oncogenes and tumor suppressors [<a href="#ref-10">10</a>]. This example illustrates that the biological question can require both gene-level and isoform-level analyses within the same study, with each analysis addressing a different aspect of the biology.

Researchers should document the primary question and the secondary questions before selecting the analysis approach. This documentation prevents scope creep and ensures that the sequencing depth and analysis resources are allocated appropriately.

Decision Point 2: Assess Available Sequencing Depth

Sequencing depth is the most practical constraint on isoform-level analysis. Gene-level analysis requires fewer reads per gene because all reads from a gene are pooled regardless of isoform identity. Isoform-level analysis distributes reads across isoforms, reducing the count per isoform and requiring higher total depth to achieve statistical power.

A practical assessment involves estimating the number of genes of interest, the expected number of isoforms per gene, and the minimum read count needed per isoform per haplotype for the chosen statistical test. For gene-level analysis, a target of 20 to 30 reads per gene per haplotype is a reasonable starting point. For isoform-level analysis, the target should be applied to each isoform, which multiplies the required depth by the number of isoforms per gene.

If the available depth is insufficient for isoform-level analysis, researchers have two options. The first option is to restrict isoform-level analysis to a subset of genes of primary interest. The second option is to proceed with gene-level analysis and acknowledge the limitation in the interpretation of results.

Decision Point 3: Evaluate Genomic Complexity

Genomic complexity affects the feasibility of both gene-level and isoform-level analysis. Polyploid organisms present additional challenges because more than two haplotypes must be distinguished and homeologous genes may share high sequence similarity. The LongPolyASE framework was developed specifically to address this gap, providing an end-to-end computational framework for allele-specific gene and isoform expression analysis in diploid and polyploid organisms [<a href="#ref-3">3</a>].

For polyploid analysis, the decision framework should include an assessment of the ploidy level, the degree of sequence similarity between homeologous genes, and the availability of a haplotype-resolved reference genome. The LongPolyASE framework includes a component for identifying syntenic gene relationships and annotation inconsistencies, which is essential for accurate analysis in polyploids [<a href="#ref-3">3</a>].

For diploid organisms, the genomic complexity assessment focuses on the density of heterozygous variants and the phase block continuity across genes of interest. Low variant density reduces the read assignment rate and may make isoform-level analysis impractical for genes in those regions.

Decision Point 4: Consider the Statistical Power Requirements

The statistical power to detect allelic imbalance depends on the read count per haplotype, the magnitude of the imbalance, and the number of biological replicates. For a single individual, the power to detect a given allelic ratio depends on the total read count. For multiple individuals, the power also depends on the between-individual variation.

A power analysis should be performed before sequencing to determine the required depth. The analysis should account for the expected read assignment rate, which is typically lower for isoform-level analysis because reads are distributed across isoforms. The power analysis should also account for the expected effect size, which is the minimum allelic imbalance that the study aims to detect.

If the power analysis indicates that the required depth is not feasible, researchers should consider whether the biological question can be addressed with a smaller set of genes or whether a different experimental design would be more efficient.

Decision Point 5: Select the Analysis Workflow

The final decision point is the selection of the specific analysis workflow. For gene-level analysis in diploid organisms, the workflow involves phasing the genome, constructing haplotype-specific references, aligning reads, and quantifying transcript expression per haplotype. For isoform-level analysis, the workflow additionally requires isoform annotation and isoform-level quantification.

For polyploid organisms, the LongPolyASE framework provides a purpose-built workflow that includes components for identifying syntenic gene relationships, discovering novel isoforms, and performing statistical testing [<a href="#ref-3">3</a>]. The framework has been applied to diploid rice, autotetraploid potato, allotetraploid rapeseed, and allooctoploid strawberry using both Oxford Nanopore and PacBio long-read RNA-seq [<a href="#ref-3">3</a>].

The workflow selection should also consider the sequencing platform. A systematic evaluation of ONT and PacBio platforms using the same single-cell cDNA libraries found that PacBio outperformed ONT in the accuracy of novel transcript identification and allele-specific gene and isoform expression [<a href="#ref-4">4</a>]. This difference is attributed to the higher sequencing quality of PacBio [<a href="#ref-4">4</a>].

Decision Matrix for Common Scenarios

Recording the Decision Process

The decision process should be documented in the analysis log. The documentation should include the biological question, the sequencing depth assessment, the genomic complexity evaluation, the power analysis results, and the selected workflow. This documentation supports reproducibility and provides context for interpreting the results.

GTN's training modules guide users through cloud-based workflow execution environments that support provenance tracking and result reproducibility.

Escalation Criteria for Decision Uncertainty

If the decision between gene-level and isoform-level analysis remains uncertain after applying this framework, researchers should escalate to specialized support. Statistical consultants can help with power analysis and experimental design. Bioinformatics core facilities can provide guidance on the feasibility of isoform-level analysis with the available data. Tool developers can provide specific guidance on the capabilities and limitations of their software.

The decision framework is not a substitute for domain expertise. Researchers with deep knowledge of their biological system may identify considerations that are not captured in this framework. The framework provides a structured starting point that can be adapted to the specific study context.

Frequently Asked Questions

What is the minimum sequencing depth for allele-specific expression analysis with long-read RNA-seq?

The minimum depth depends on the number of genes of interest and the magnitude of allelic imbalance to be detected. A practical starting point is 20 to 30 reads per gene per haplotype for genes of interest. This depth provides reasonable power for detecting moderate imbalance. Isoform-level analysis requires higher depth because reads are distributed across isoforms. Perform a power analysis before sequencing to determine the required depth for your specific study design.

Can I use short-read RNA-seq data for allele-specific expression analysis instead of long reads?

Short-read RNA-seq can be used for gene-level ASE analysis when reads overlap heterozygous variants. However, short reads cannot reliably distinguish isoforms that share exons, limiting isoform-level analysis. Long reads provide the advantage of spanning multiple variants and capturing complete transcript structures, enabling both allele-specific expression and allele-specific transcript structure analysis [<a href="#ref-1">1</a>]. The choice depends on whether isoform-level resolution is needed for the research question.

How do I choose between Oxford Nanopore and PacBio for ASE analysis?

PacBio generally provides higher base-level accuracy than ONT, which translates to more accurate novel transcript identification and allele-specific gene and isoform expression [<a href="#ref-4">4</a>]. ONT offers advantages in throughput and cost for some applications. The choice should be guided by the research question, the required accuracy, and the available budget. For studies where novel isoform identification is critical, PacBio may be worth the additional cost.

What is the difference between gene-level and isoform-level allele-specific expression?

Gene-level ASE compares the total expression from each haplotype for a gene. Isoform-level ASE compares the expression of each isoform between haplotypes. A gene can show balanced total expression while having imbalanced isoform usage, such as one haplotype producing predominantly one isoform and the other haplotype producing a different isoform. Long-read data is particularly suited to isoform-level analysis because full-length reads capture complete transcript structures.

How do I handle polyploid organisms in ASE analysis?

Polyploid analysis requires specialized tools because more than two haplotypes must be distinguished and homeologous genes may share high sequence similarity. The LongPolyASE framework provides an end-to-end computational framework for allele-specific gene and isoform expression analysis in diploid and polyploid organisms [<a href="#ref-3">3</a>]. It has been applied to autotetraploid potato, allotetraploid rapeseed, and allooctoploid strawberry, demonstrating its utility for complex genomes.

What causes low read assignment to haplotypes and how can I fix it?

Low read assignment can result from low heterozygosity, short phase blocks, or high sequencing error rates. Check the heterozygosity of the sample, examine phase block lengths for genes of interest, and verify that phasing was performed correctly. Increasing sequencing depth for phasing or using a haplotype-resolved reference genome can improve assignment rates. Genes with low assignment rates should be excluded from ASE analysis or interpreted with caution.

How do I detect and reduce reference bias in ASE analysis?

Reference bias occurs when reads from one haplotype align preferentially to the reference genome. This bias can produce false allelic imbalance. Using a haplotype-resolved reference genome reduces bias compared to using a linear reference. Test for bias by examining the allelic balance in genes known to have balanced expression. If bias is detected, consider using a different alignment approach or a variant-substituted reference.

What validation is needed before reporting ASE findings?

ASE findings should be validated with orthogonal methods before drawing strong conclusions. Validation approaches include quantitative PCR with allele-specific primers, targeted sequencing with different technology, or functional assays. The need for validation depends on the significance of the finding and its potential impact. Findings used for clinical decisions, breeding decisions, or high-impact publications should be validated. Exploratory findings may not require immediate validation.

Related Bioinformatics Guides

• RNA Sequencing Data Analysis: From Raw Reads to Differential Expression
• RNA-Seq Data Analysis Workflow: From Raw Reads to Insights
• RNA-Seq Data Analysis in Galaxy: A User-Friendly Platform
• RNA-Seq vs Microarray: Choosing the Right Gene Expression Profiling Platform
• Full-Length Transcript Sequencing: Unraveling Isoform Diversity with Long Reads

References and Further Reading

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