# How to Quantify Transcript Isoforms from Nanopore RNA-Seq Data: A Comparison of Salmon, Flair, and bambu

## Direct Answer and Scope

Oxford Nanopore RNA sequencing produces full-length transcript reads that can resolve isoform structure and expression in ways short-read platforms cannot. The practical question for a bioinformatics practitioner is which quantification tool to run after basecalling and alignment. This article compares three widely used approaches: Salmon, Flair, and bambu. Salmon is a lightweight alignment-free quantifier originally built for short reads but adapted for long reads. Flair is a long-read-specific pipeline that corrects isoforms and quantifies them from nanopore data. bambu is an R/Bioconductor package that uses reference annotations to estimate isoform expression directly from long-read alignments. Each tool makes different assumptions about error profiles, annotation completeness, and the need for transcript discovery. The choice among them changes the accuracy of isoform-level counts, the runtime required, and the interpretability of the output. This comparison uses published benchmark evidence from the Long-read RNA-Seq Genome Annotation Assessment Project (LRGASP) and related studies to give concrete decision criteria for researchers working with nanopore data.

The intended reader is a biology student, researcher, or laboratory professional who has generated or plans to generate nanopore RNA-seq data and needs to select a quantification strategy. The practical outcome is a workflow decision: which tool to run for a given project goal, what controls to include, what records to keep, and when to escalate to a more complex approach. The comparison assumes the reader has basic familiarity with RNA-seq analysis but does not require prior experience with long-read quantification tools.

## At a Glance: Tool Comparison for Nanopore Isoform Quantification

The table below summarizes the key differences among Salmon, Flair, and bambu for nanopore RNA-seq isoform quantification. The evidence base draws from the LRGASP consortium evaluation, which generated over 427 million long-read sequences from cDNA and direct RNA datasets across human, mouse, and manatee species using different protocols and sequencing platforms.

| Feature | Salmon | Flair | bambu |
|---------|--------|-------|-------|
| Primary input | Raw or aligned reads | Aligned reads (SAM/BAM) | Aligned reads (BAM) with reference annotation |
| Transcript discovery | No, quantifies against provided transcriptome | Yes, corrects and discovers isoforms | Yes, uses annotation to model novel isoforms |
| Error handling | Uses error models for long reads | Explicit error correction of splice junctions | Uses annotation and read counts to filter artifacts |
| Reference requirement | Transcriptome FASTA | Genome FASTA and annotation GTF | Genome FASTA and annotation GTF |
| Runtime profile | Fast, alignment-free | Moderate, requires multiple steps | Moderate, R-based processing |
| Best use case | Well-annotated genomes, quick quantification | Novel isoform discovery with correction | Balanced discovery and quantification with annotation |
| Output format | Quantification tables, transcript-level counts | Isoform GTF and quantification | Transcript-level expression estimates |
| Reproducibility support | Command-line, scriptable | Pipeline-oriented, scriptable | Bioconductor workflow, versioned |

The LRGASP study found that libraries with longer and more accurate sequences produce more accurate transcripts than libraries with increased read depth alone, while greater read depth improved quantification accuracy. This finding directly affects tool choice: if your nanopore library has short or error-prone reads, no quantification tool will fully compensate, and you should prioritize library preparation improvements before tool selection. In well-annotated genomes, reference-based tools demonstrated the best performance in the LRGASP evaluation, which supports the use of bambu or Flair in reference-guided mode for human and mouse data.

## Context: Why Isoform Quantification from Nanopore Data Differs from Short-Read Approaches

### The Transcriptome Complexity Problem

Standard short-read RNA-seq measures gene expression by counting fragments that map to exons or exon junctions. This approach works well for gene-level quantification but struggles with isoform-level resolution because short reads cannot span full-length transcripts. The transcriptome consists of variable-length transcripts and multiple alternatively spliced isoforms for most genes, and many isoforms share exons, making assignment of short reads to specific isoforms ambiguous. Long-read sequencing technologies from Pacific Biosciences and Oxford Nanopore Technologies capture full-length cDNA molecules, which allows direct observation of exon connectivity and isoform structure. The shift toward long-read sequencing for transcriptome characterization is driven by increases in throughput and decreases in cost, making these platforms attractive for de novo transcriptome assembly, isoform expression quantification, and in-depth RNA species analysis.

### What Nanopore Reads Provide and What They Do Not

Nanopore sequencing produces reads that can span entire transcripts, but the raw error rate is higher than short-read platforms. The relatively high error rate of current long-read sequencing platforms poses a major challenge for isoform discovery and quantification. Errors in homopolymer regions and base modifications can create false splice junctions or miscalled bases that complicate alignment and quantification. Tools like ESPRESSO were developed specifically to address this challenge by jointly considering alignments of all long reads aligned to a gene and using error profiles of individual reads to improve the identification of splice junctions and the discovery of corresponding transcript isoforms. The same error considerations apply to Salmon, Flair, and bambu, though each handles errors differently.

### The Role of Reference Annotations

The availability and quality of reference annotations strongly influence tool performance. The LRGASP evaluation showed that in well-annotated genomes, tools based on reference sequences demonstrated the best performance. For human and mouse data, comprehensive annotations from NCBI and Ensembl provide a strong foundation for reference-guided quantification. For non-model organisms or poorly annotated genomes, reference-free approaches may be necessary, but the LRGASP study advises incorporating additional orthogonal data and replicate samples when aiming to detect rare and novel transcripts or when using reference-free approaches. This context matters for tool selection because Salmon requires a transcriptome FASTA, while Flair and bambu can work with genome and annotation files.

## Core Principles of Isoform Quantification from Nanopore Data

### Quantification Is Distinct from Discovery

Isoform quantification assigns expression values to known or newly identified transcript models. Isoform discovery identifies the set of transcript models present in a sample. These are related but distinct tasks. Salmon quantifies against a provided transcriptome and does not discover new isoforms. Flair performs both correction and discovery, then quantifies the resulting isoform set. bambu uses the reference annotation as a starting point, models novel isoforms that are supported by the data, and quantifies all transcripts in a unified framework. The choice of tool depends on whether your project requires discovery, quantification, or both.

### Read Depth versus Read Length Tradeoffs

The LRGASP consortium evaluation revealed a critical tradeoff: libraries with longer and more accurate sequences produce more accurate transcripts than those with increased read depth, whereas greater read depth improved quantification accuracy. This means that a library with 10 million long, high-quality reads may yield better isoform discovery than a library with 30 million short, error-prone reads. For quantification accuracy, however, more reads help even if they are shorter. When planning a nanopore experiment, you should consider whether your primary goal is discovery or quantification and design library preparation accordingly. If you need both, you may need to balance read length and depth.

### The Importance of Replicates

The LRGASP study specifically advises incorporating additional orthogonal data and replicate samples when detecting rare or novel transcripts or when using reference-free approaches. Biological replicates capture inter-individual and inter-sample variation, while technical replicates capture platform and library preparation variation. For isoform quantification, replicates are essential for statistical testing of differential isoform expression. A single nanopore run can identify isoforms, but it cannot support claims about differential expression without replication. This principle applies regardless of which quantification tool you choose.

## Practical Workflow: From Raw Nanopore Reads to Isoform Quantification

### Step 1: Basecalling and Quality Control

The first stage in any nanopore RNA-seq workflow is basecalling, which converts raw electrical signal data into nucleotide sequences. Basecalling quality directly affects downstream quantification accuracy. After basecalling, run quality control to assess read length distribution, read quality scores, and the presence of adapter contamination. The Galaxy Training Network provides accessible workflow training for quality control and analysis steps, and the EMBL-EBI Training portal offers structured learning pathways for bioinformatics data analysis. These resources are useful for researchers who need to build or verify their analysis pipelines.

### Step 2: Alignment to the Reference Genome

For Flair and bambu, alignment to a reference genome is required. The choice of aligner matters because splice-aware alignment is necessary for RNA-seq data. Minimap2 is a common choice for long-read alignment, though other splice-aware aligners exist. The alignment step produces a BAM file that contains the genomic coordinates and splicing information for each read. For Salmon, alignment is optional because Salmon can operate in alignment-free mode using k-mer counting against a transcriptome index. However, if you already have aligned data, Salmon can also accept aligned reads as input.

### Step 3: Transcript Model Construction or Correction

Flair corrects isoforms by using read alignments to identify and correct splice junctions, then collapses reads into isoform models. This step is computationally intensive but necessary for accurate isoform discovery from error-prone nanopore data. bambu constructs transcript models by comparing read alignments to the reference annotation and identifying novel isoforms that are consistently supported by multiple reads. Salmon does not perform this step because it quantifies against a provided transcriptome.

### Step 4: Quantification

The quantification step assigns read counts or abundance estimates to each transcript model. Salmon uses a lightweight approach that estimates transcript-level abundance from k-mer counts. Flair counts reads that support each corrected isoform. bambu uses an expectation-maximization algorithm to estimate transcript expression from read alignments, accounting for multi-mapping reads and isoform ambiguity. The output of each tool is a table of transcript-level expression estimates that can be used for downstream analysis.

### Step 5: Quality Assessment of Quantification Results

After quantification, assess the results for quality. Check the number of transcripts detected, the distribution of expression values, and the concordance with expected biology. If you have spike-in controls or known expression ratios, compare the quantified values to the expected values. The Bioconductor project provides documentation and workflows for reproducible genomic analysis, including quality assessment steps for RNA-seq data.

## Tool-Specific Workflow Details

### Salmon for Nanopore Data

Salmon was originally designed for short-read RNA-seq but has been adapted for long-read data. The tool operates in alignment-free mode by building an index of k-mers from the transcriptome and then counting k-mers in the reads. For nanopore data, Salmon can use error models that account for the higher error rate of long reads. The main advantage of Salmon is speed: it does not require alignment and can process large datasets quickly. The main limitation is that it cannot discover new isoforms. If your transcriptome annotation is incomplete, Salmon will miss novel isoforms and may misassign reads from novel isoforms to known isoforms with similar sequences.

To use Salmon for nanopore data, you need a transcriptome FASTA file. This file can come from a reference annotation or from a custom transcriptome assembled from your data. If you use a reference transcriptome, ensure it matches the genome version used for alignment and annotation. Salmon requires a decoy sequence file to handle reads that map to genomic regions outside annotated transcripts. The decoy file should contain the genome sequence to prevent spurious alignments to unannotated regions.

### Flair for Nanopore Data

Flair is a pipeline specifically designed for long-read RNA-seq data. It takes aligned reads as input and performs three main steps: correction, collapse, and quantification. The correction step uses read alignments to identify and correct splice junctions, reducing the impact of sequencing errors on isoform discovery. The collapse step groups reads into isoform models based on their splice patterns. The quantification step counts reads supporting each isoform.

Flair requires a genome FASTA file and an annotation GTF file. The annotation is used to identify known isoforms and to provide context for novel isoform discovery. Flair can also use a short-read dataset to help correct isoforms, though this is optional. The pipeline is scriptable and can be integrated into larger workflows. The main advantage of Flair is its explicit error correction, which is valuable for nanopore data with high error rates. The main limitation is that it requires multiple steps and is more computationally intensive than Salmon.

### bambu for Nanopore Data

bambu is an R/Bioconductor package that estimates isoform expression from long-read alignments. It uses the reference annotation as a starting point and models novel isoforms that are supported by the data. bambu applies a machine learning approach to distinguish real isoforms from artifacts caused by sequencing errors. The package provides functions for filtering low-confidence isoforms and for visualizing isoform structures.

bambu requires a genome FASTA file, an annotation GTF file, and aligned reads in BAM format. The package can process multiple samples simultaneously and provides a unified output table of transcript-level expression estimates. bambu is well-documented within the Bioconductor framework, which supports reproducible genomic analysis. The main advantage of bambu is its balance between discovery and quantification: it can identify novel isoforms while providing accurate expression estimates. The main limitation is that it requires R and may have a steeper learning curve for researchers unfamiliar with R.

## Options and Tradeoffs: Choosing the Right Tool for Your Project

### Project Goal: Quantification of Known Isoforms

If your project goal is to quantify expression of known isoforms in a well-annotated genome, Salmon is the most efficient choice. The LRGASP evaluation found that reference-based tools performed best in well-annotated genomes. Salmon provides fast, accurate quantification without the computational overhead of isoform discovery. This approach is suitable for projects that compare isoform expression across conditions in human or mouse samples with comprehensive annotations.

### Project Goal: Novel Isoform Discovery

If your project goal is to discover novel isoforms, Flair or bambu are appropriate choices. Flair provides explicit error correction, which is valuable for identifying rare or novel isoforms from error-prone nanopore data. bambu uses the reference annotation to model novel isoforms and provides a unified framework for discovery and quantification. The ESPRESSO study demonstrated that tools jointly considering alignments of all long reads to a gene and using error profiles of individual reads improve the identification of splice junctions and the discovery of corresponding transcript isoforms. This principle applies to Flair and bambu, though each implements it differently.

### Project Goal: Balanced Discovery and Quantification

If your project requires both discovery and quantification, bambu offers the most integrated approach. The package models novel isoforms and quantifies all transcripts in a single framework, reducing the risk of inconsistencies between discovery and quantification steps. Flair also provides both functions but requires separate steps for correction, collapse, and quantification. The choice between bambu and Flair may depend on your familiarity with R and your preference for integrated versus modular pipelines.

### Computational Resources

Salmon is the fastest option because it does not require alignment. Flair and bambu require alignment and additional processing steps, which increases runtime and memory usage. If you have limited computational resources or need to process many samples quickly, Salmon may be the most practical choice. If you have access to a high-performance computing cluster, the additional runtime of Flair or bambu may be acceptable.

### Annotation Quality

The quality of your reference annotation is a critical factor. For well-annotated genomes like human and mouse, reference-based tools perform well. For poorly annotated genomes, reference-free approaches may be necessary, but the LRGASP study advises incorporating additional orthogonal data and replicate samples in these cases. If your organism has a poor annotation, consider whether you need to generate a custom transcriptome or use a reference-free tool before selecting a quantification approach.

## Observations and Measurements: What to Record During Analysis

### Read-Level Metrics

Record the number of reads generated, the read length distribution, and the read quality scores. These metrics affect downstream quantification accuracy. The LRGASP study found that libraries with longer and more accurate sequences produce more accurate transcripts than those with increased read depth. If your read length distribution is short or your quality scores are low, you should expect reduced isoform discovery accuracy regardless of the quantification tool used.

### Alignment Metrics

Record the alignment rate, the number of reads that map to the reference genome, and the number of reads that map to multiple locations. Multi-mapping reads create ambiguity in isoform quantification because they cannot be uniquely assigned to a single transcript. Tools like bambu use expectation-maximization to handle multi-mapping reads, but the presence of many multi-mapping reads reduces the confidence of isoform-level estimates.

### Quantification Metrics

Record the number of transcripts detected, the number of novel isoforms identified, and the distribution of expression values. Compare these metrics across samples to identify outliers or technical failures. If a sample has an unusually low number of detected transcripts, this may indicate a library preparation problem or a sequencing failure.

### Tool Version and Parameters

Record the version of each tool used and the parameters applied. Tool versions can change behavior, and parameter choices affect results. The Bioconductor project emphasizes reproducible genomic analysis, which requires documenting the software environment and analysis parameters. The nf-core documentation provides standards for community pipeline usage and configuration, which can help you structure reproducible workflows.

## Records and Documentation for Reproducibility

### Analysis Logs

Maintain a log of all analysis steps, including software versions, parameters, and input files. This log should be sufficient for another researcher to reproduce your analysis. The Carpentries lessons provide foundational training in computing, data, shell, Git, and programming, which includes best practices for documenting analysis workflows.

### Data Management

Store raw sequencing data, aligned reads, and quantification results in a structured directory hierarchy. Use descriptive file names that include sample identifiers, tool names, and version numbers. The NCBI Data Resources provide official descriptions of databases and search systems for sequence data, which can help you organize and archive your data for public deposition.

### Version Control

Use version control for analysis scripts and configuration files. Git is a common tool for this purpose, and the Carpentries lessons include training in Git for version control. Version control allows you to track changes to your analysis and to revert to previous versions if needed.

### Containerization

Consider using containers to encapsulate your analysis environment. Containers ensure that software dependencies are consistent across different computing systems. The nf-core documentation provides guidance on container usage and configuration for reproducible workflows. Containers are particularly useful for long-read analysis pipelines that depend on specific versions of aligners and quantification tools.

## Common Failure Patterns and How to Address Them

### Low Alignment Rate

A low alignment rate indicates that many reads do not map to the reference genome. This can result from adapter contamination, poor basecalling quality, or a mismatch between the read sequences and the reference genome. Check the read quality metrics and the adapter content. If the reference genome is incomplete or divergent from your sample, consider using a different reference or a reference-free approach.

### Excessive Novel Isoform Calls

If Flair or bambu identifies an implausibly large number of novel isoforms, this may indicate that sequencing errors are being interpreted as real splice variants. The ESPRESSO study noted that the relatively high error rate of current long-read sequencing platforms poses a major challenge for isoform discovery. Tools that use error profiles of individual reads can reduce false novel isoform calls. Check the read quality and consider applying stricter filtering parameters.

### Inconsistent Quantification Across Replicates

If biological replicates show highly variable isoform-level expression, this may indicate technical variation or a failure in library preparation. The LRGASP study advises incorporating replicate samples when detecting rare or novel transcripts. Check the read depth and read length distribution across replicates. If one replicate has substantially fewer reads or shorter reads, this may explain the inconsistency.

### Runtime or Memory Failures

Flair and bambu can be computationally intensive, especially for large datasets. If the analysis fails due to memory limits, consider processing chromosomes separately or using a high-performance computing cluster. Salmon is less computationally demanding and may be a better choice if computational resources are limited.

## Limitations of Each Tool and of Nanopore Quantification in General

### Salmon Limitations

Salmon cannot discover novel isoforms. If your transcriptome annotation is incomplete, Salmon will miss novel isoforms and may misassign reads from novel isoforms to known isoforms with similar sequences. Salmon also requires a transcriptome FASTA file, which may not be available for non-model organisms.

### Flair Limitations

Flair requires multiple processing steps and is more computationally intensive than Salmon. The error correction step can be slow for large datasets. Flair also depends on the quality of the reference annotation for context, and its performance may degrade for poorly annotated genomes.

### bambu Limitations

bambu requires R and may have a steeper learning curve for researchers unfamiliar with R. The package uses a machine learning approach to filter novel isoforms, which requires training data and may not perform well for organisms with unusual transcriptome features. bambu also requires a reference annotation, which limits its use for de novo transcriptome analysis.

### General Limitations of Nanopore Quantification

Nanopore sequencing has a higher error rate than short-read sequencing, which affects isoform discovery and quantification accuracy. The LRGASP study found that libraries with longer and more accurate sequences produce more accurate transcripts than those with increased read depth. This means that library preparation quality is a critical factor in quantification accuracy. Additionally, nanopore RNA-seq cannot currently match the depth of short-read sequencing at the same cost, which may limit the detection of low-abundance isoforms.

## Quality and Welfare Controls in the Analysis Context

### Spike-In Controls

Spike-in controls are synthetic RNA molecules with known sequences and concentrations that can be added to your library before sequencing. These controls provide a ground truth for assessing quantification accuracy. Compare the quantified expression of spike-in controls to the expected concentrations to identify systematic biases in your workflow.

### Technical Replicates

Technical replicates are repeated measurements from the same biological sample. They capture variation introduced by library preparation and sequencing. The LRGASP study advises incorporating replicate samples when detecting rare or novel transcripts. Technical replicates can help you distinguish technical noise from biological variation.

### Orthogonal Validation

The LRGASP study recommends incorporating additional orthogonal data when detecting rare or novel transcripts. Orthogonal data can include short-read RNA-seq, quantitative PCR, or proteomics data. If a novel isoform is detected by nanopore sequencing but not by an orthogonal method, this may indicate a false positive.

### Professional Escalation Criteria

If you observe any of the following, consider escalating to a more experienced bioinformatician or a specialized service provider: persistent low alignment rates that cannot be resolved by quality filtering, implausible isoform structures that suggest systematic errors, or inconsistent results across replicates that cannot be explained by technical variation. Additionally, if you are working with a non-model organism and need to build a custom transcriptome, consider consulting with a specialist in genome annotation.

## Safety and Regulatory Context for Data Handling

### Data Privacy and Consent

RNA-seq data derived from human samples may contain identifiable genetic information. Ensure that your data collection and analysis comply with institutional review board requirements and data protection regulations. The NCBI Data Resources provide official descriptions of databases and search systems, including guidance on data deposition and access controls.

### Data Deposition

Many journals require that sequencing data be deposited in public databases. The NCBI provides databases for sequence data deposition, including the Sequence Read Archive. Before depositing data, ensure that you have the necessary consent and that the data are appropriately de-identified.

### Software Licensing

Some bioinformatics tools have specific licensing terms. Check the license for each tool you use to ensure compliance. The Bioconductor project provides open-source packages with permissive licenses, while other tools may have more restrictive terms.

## Practical Implementation Steps for Your First Nanopore Quantification Project

### Step 1: Define Your Project Goal

Write a clear statement of your project goal. Are you quantifying known isoforms, discovering novel isoforms, or both? This decision determines which tool to use and how to design your library preparation.

### Step 2: Assess Your Annotation Quality

Check the availability and quality of reference annotations for your organism. For well-annotated genomes, reference-based tools are appropriate. For poorly annotated genomes, consider whether you need a reference-free approach or additional orthogonal data.

### Step 3: Select Your Tool

Based on your project goal and annotation quality, select Salmon, Flair, or bambu. Use the At a Glance table to compare features and choose the tool that best matches your needs.

### Step 4: Prepare Your Analysis Environment

Install the required software and verify that it runs correctly. Use containers or version control to ensure reproducibility. The nf-core documentation provides standards for community pipeline usage and configuration.

### Step 5: Run Quality Control on Your Reads

Assess read length distribution, quality scores, and adapter content. If quality is poor, consider whether library preparation needs to be repeated before proceeding with quantification.

### Step 6: Align or Index

For Flair and bambu, align reads to the reference genome. For Salmon, build a transcriptome index. Record the parameters used for each step.

### Step 7: Run Quantification

Run the selected quantification tool. Record the output files and the parameters used. Check the output for quality indicators such as the number of transcripts detected and the distribution of expression values.

### Step 8: Validate Your Results

Compare your results to spike-in controls, technical replicates, or orthogonal data. If discrepancies are observed, investigate the cause before proceeding with downstream analysis.

### Step 9: Document and Archive

Document all analysis steps, software versions, and parameters. Archive raw data, aligned reads, and quantification results in a structured directory. Consider depositing data in a public database if required by your journal or funding agency.

## Common Failure Patterns in Tool Selection and Use

### Choosing Salmon for Novel Isoform Discovery

A common mistake is using Salmon when the project goal includes novel isoform discovery. Salmon cannot discover new isoforms and will miss transcripts that are not in the provided transcriptome. If you suspect your annotation is incomplete, use Flair or bambu instead.

### Using an Outdated Annotation

Reference annotations are updated regularly as new evidence accumulates. Using an outdated annotation can lead to missed isoforms or incorrect quantification. Check the annotation version and update it if a newer version is available.

### Ignoring Read Quality

The LRGASP study found that read length and accuracy are more important than read depth for transcript discovery. If your reads are short or error-prone, no quantification tool will fully compensate. Prioritize library preparation quality over read depth.

### Failing to Document Parameters

Quantification results depend on the parameters used. Failing to document parameters makes it impossible to reproduce the analysis or to compare results across studies. Use version control and analysis logs to document all parameters.

## Interpretation Limits and Reporting Standards

### What Quantification Numbers Mean

Transcript-level expression estimates from nanopore data are relative abundance measures, not absolute molecule counts. They reflect the proportion of reads assigned to each transcript, adjusted for transcript length and sequencing bias. Do not interpret these numbers as absolute copy numbers without calibration.

### Reporting Requirements

When reporting isoform quantification results, include the tool version, parameters, reference annotation version, and quality metrics. This information allows other researchers to evaluate the reliability of your results. The EMBL-EBI Training portal provides guidance on reporting standards for bioinformatics analyses.

### Limitations of Cross-Study Comparisons

Isoform quantification results are difficult to compare across studies that use different tools, parameters, or annotation versions. If you compare your results to published data, ensure that the analysis methods are compatible. The LRGASP study provides a benchmark for current practices and offers direction for future method development in transcriptome analysis.

## Decision Framework: Matching Project Constraints to Tool Selection

Selecting between Salmon, Flair, and bambu requires more than matching a tool to a project goal. Practical constraints around annotation completeness, computational capacity, and validation requirements often override the ideal tool choice. This framework provides a structured decision process based on evidence from the LRGASP consortium evaluation and related long-read benchmarking studies.

### Constraint 1: Annotation Completeness and Quality

The LRGASP evaluation found that in well-annotated genomes, tools based on reference sequences demonstrated the best performance. This finding creates a clear decision branch. For human and mouse data with comprehensive annotations from NCBI or Ensembl, reference-guided quantification with bambu or Flair produces the most reliable results. For non-model organisms or genomes with sparse annotations, the same study advises incorporating additional orthogonal data and replicate samples when using reference-free approaches.

Assess your annotation before selecting a tool. Check the number of annotated transcripts per gene, the date of the annotation release, and whether the annotation includes alternative isoforms or only representative transcripts. An annotation that lists one transcript per gene will not support meaningful isoform quantification regardless of tool choice. If your annotation lacks isoform diversity, you must either generate a custom transcriptome from your data or accept gene-level resolution as your primary output.

### Constraint 2: Computational Capacity and Throughput Requirements

Salmon operates without alignment and provides the fastest path from raw reads to quantification tables. This advantage matters when processing large cohorts or when working within limited computing environments. Flair requires multiple sequential steps including correction, collapse, and quantification, which increases runtime substantially. bambu operates within R and processes aligned reads through an expectation-maximization algorithm, with memory usage scaling with alignment file size.

Estimate your computational budget before committing to a tool. A project with 50 samples and a 48-hour processing window cannot accommodate Flair or bambu unless you have cluster access. Conversely, a project with five samples and access to high-performance computing can absorb the additional runtime of discovery-capable tools. The nf-core documentation provides standards for pipeline configuration that can help you estimate resource requirements before launching an analysis.

### Constraint 3: Validation Requirements and Downstream Analysis Needs

The LRGASP study specifically recommends incorporating orthogonal data and replicate samples when detecting rare or novel transcripts. This recommendation has direct implications for tool selection. If your project requires validation of novel isoforms through orthogonal methods, you need a tool that produces explicit isoform models you can test. Flair outputs a GTF file of corrected isoforms that can be directly compared to short-read data or PCR validation results. bambu provides transcript models within its R object structure, which requires additional steps to export for external validation.

Salmon produces quantification tables without transcript models, making it unsuitable for projects that require validation of specific isoform structures. If your downstream analysis includes differential isoform expression testing, confirm that your chosen tool's output format is compatible with your statistical analysis package. The Bioconductor project provides documentation for integrating quantification outputs into differential expression workflows.

### Constraint 4: Error Tolerance and Read Quality Profile

The ESPRESSO study demonstrated that tools jointly considering alignments of all long reads to a gene and using error profiles of individual reads improve splice junction identification and isoform discovery. This evidence supports the use of error-aware tools when your nanopore data has elevated error rates. Flair performs explicit error correction of splice junctions. bambu applies machine learning to distinguish real isoforms from artifacts. Salmon relies on error models but does not correct individual reads.

Evaluate your read quality before selecting a tool. If your median read accuracy falls below typical nanopore thresholds or if you observe homopolymer errors that affect splice junction calls, prioritize tools with explicit error handling. If your reads are high quality with minimal error, the error correction overhead of Flair may not justify the additional runtime.

### Constraint 5: Reproducibility and Documentation Requirements

The Carpentries lessons provide foundational training in computing, data, shell, Git, and programming, which includes best practices for documenting analysis workflows. Your tool choice affects how easily you can meet reproducibility standards. Salmon runs as a single command with a defined index and parameter set, making it straightforward to document. Flair requires multiple commands with intermediate files, increasing the documentation burden. bambu operates within R scripts that can be version-controlled and shared through Bioconductor workflows.

Consider your reporting obligations before selecting a tool. Journals increasingly require detailed methods sections that include software versions, parameters, and analysis steps. The EMBL-EBI Training portal provides guidance on reporting standards for bioinformatics analyses. If your project requires public data deposition, the NCBI Data Resources provide official descriptions of databases and search systems for sequence data, including requirements for analysis metadata.

## Practical Assessment Checklist for Tool Selection

Use this checklist to document your decision process and to create a record that supports reproducibility.

### Step 1: Document Your Project Goal

Write a specific statement of whether your project requires quantification of known isoforms, discovery of novel isoforms, or both. Record this statement in your analysis log. The LRGASP evaluation found that libraries with longer and more accurate sequences produce more accurate transcripts than those with increased read depth, while greater read depth improved quantification accuracy. Your goal determines whether you prioritize read length or read depth in library preparation and which tool characteristics matter most.

### Step 2: Assess Your Annotation

Record the annotation source, version, and date of download. Count the number of transcripts per gene for a sample of genes to assess isoform representation. If your annotation has fewer than two transcripts per gene for most genes, isoform quantification will be limited regardless of tool choice. The NCBI Data Resources provide official descriptions of annotation databases and search systems that can help you verify annotation completeness.

### Step 3: Evaluate Your Read Quality Profile

Generate read length distribution and quality score summaries from your basecalled data. Record the median read length, the N50 read length, and the proportion of reads passing your quality threshold. The LRGASP study found that read length and accuracy are more important than read depth for transcript discovery. If your reads are short or error-prone, address library preparation before selecting a quantification tool.

### Step 4: Estimate Computational Requirements

Calculate the number of samples, the expected read depth per sample, and the available computing resources. Estimate runtime and memory requirements for each candidate tool based on these parameters. The nf-core documentation provides standards for pipeline usage and configuration that can help you structure resource estimates.

### Step 5: Select Your Primary and Backup Tools

Choose a primary tool based on your project goal, annotation quality, and computational capacity. Select a backup tool that can validate your primary results. The LRGASP study advises incorporating additional orthogonal data when detecting rare or novel transcripts. Running a second quantification tool on the same data provides a form of internal validation that can identify systematic biases in either tool.

### Step 6: Define Success Criteria Before Running the Analysis

Write specific criteria for acceptable results before you run the quantification. These criteria should include minimum alignment rates, expected numbers of detected transcripts, and concordance thresholds between replicates. The Galaxy Training Network provides accessible workflow training that can help you define quality thresholds for your analysis.

### Step 7: Record the Decision and Rationale

Document your tool selection, the constraints that influenced your choice, and the evidence supporting your decision. This record supports reproducibility and provides context for interpreting your results. The Carpentries lessons include training in Git for version control, which can help you track changes to your analysis decisions over time.

## Records and Measurements for Tool Comparison

Maintain structured records that allow you to compare tool performance across projects and to identify systematic differences between tools.

### Runtime and Resource Logs

Record the wall-clock time, peak memory usage, and CPU hours for each tool run. These metrics inform future project planning and help you estimate resource requirements for larger datasets. The nf-core documentation provides standards for pipeline configuration that include resource reporting.

### Quantification Concordance Metrics

When running multiple tools on the same data, record the correlation between transcript-level expression estimates. High concordance between tools increases confidence in your results. Low concordance may indicate that one tool is misassigning reads or that your annotation does not adequately represent the transcriptome.

### Novel Isoform Validation Records

If you use Flair or bambu for novel isoform discovery, record the number of novel isoforms identified, the read support for each isoform, and any orthogonal validation results. The ESPRESSO study demonstrated that tools using error profiles of individual reads improve splice junction identification and isoform discovery. Your records should capture whether your novel isoforms meet the evidence thresholds recommended in the literature.

### Version and Parameter Documentation

Record the exact version of each tool, the reference annotation version, and all parameters used. The Bioconductor project emphasizes reproducible genomic analysis, which requires documenting the software environment and analysis parameters. Version differences can change quantification results, and parameter choices affect sensitivity and specificity.

## Common Failure Patterns in Tool Selection

### Selecting a Tool Before Assessing Annotation Quality

A frequent error is choosing a quantification tool before evaluating whether the reference annotation supports isoform-level analysis. If your annotation lacks isoform diversity, Salmon will produce misleading results because it cannot discover new isoforms. Flair and bambu can identify novel isoforms, but they require sufficient read support and may produce noisy results with sparse annotations.

### Assuming More Reads Always Improve Results

The LRGASP evaluation found that libraries with longer and more accurate sequences produce more accurate transcripts than those with increased read depth. This finding contradicts the common assumption that sequencing more reads will solve quantification problems. If your reads are short or error-prone, adding depth will not compensate for poor read quality. Address library preparation before increasing sequencing depth.

### Ignoring Computational Constraints Until Runtime Failure

Flair and bambu can fail on large datasets when memory limits are exceeded. Estimate resource requirements before launching an analysis instead of discovering limitations during runtime. The nf-core documentation provides guidance on resource estimation and pipeline configuration that can prevent these failures.

### Failing to Validate Novel Isoform Calls

The LRGASP study advises incorporating additional orthogonal data and replicate samples when detecting rare or novel transcripts. A common failure is reporting novel isoforms without orthogonal validation. If a novel isoform is detected by nanopore sequencing but not by an orthogonal method, this may indicate a false positive caused by sequencing errors.

### Using Inconsistent Parameters Across Replicates

Quantification results depend on the parameters used. If you run different parameters for different replicates, you cannot compare results across samples. Document parameters and use identical settings for all samples in a project. The Carpentries lessons provide training in version control and documentation practices that support consistent analysis.

## Professional Escalation Criteria

Escalate to a more experienced bioinformatician or specialized service provider when you encounter any of the following situations.

### Persistent Low Alignment Rates

If alignment rates remain below acceptable thresholds after quality filtering and parameter adjustment, the problem may lie in the reference genome or the library preparation. A specialist can assess whether the reference is appropriate for your sample or whether the library preparation protocol needs modification.

### Implausible Isoform Structures

If Flair or bambu identifies isoforms with structures that contradict known biology, such as exons that do not match the genome assembly or splice junctions that violate consensus sequences, this may indicate systematic errors. The ESPRESSO study noted that the relatively high error rate of current long-read sequencing platforms poses a major challenge for isoform discovery. A specialist can help you distinguish real novel isoforms from artifacts.

### Inconsistent Results Across Replicates

If biological replicates show highly variable isoform-level expression that cannot be explained by technical variation, the problem may lie in the experimental design or the analysis pipeline. The LRGASP study advises incorporating replicate samples when detecting rare or novel transcripts. A specialist can help you identify whether the variation is biological or technical.

### Non-Model Organism Annotation Challenges

If you are working with a non-model organism and need to build a custom transcriptome, consider consulting with a specialist in genome annotation. The LRGASP study found that reference-based tools performed best in well-annotated genomes, which means non-model organisms require different approaches. A specialist can help you generate a suitable transcriptome reference or select an appropriate reference-free strategy.

### Regulatory or Data Deposition Requirements

If your project involves human data or requires public data deposition, ensure that your analysis complies with institutional review board requirements and data protection regulations. The NCBI Data Resources provide official descriptions of databases and search systems for sequence data, including guidance on data deposition and access controls. A specialist can help you navigate these requirements.

## Frequently Asked Questions

### What is the main difference between Salmon, Flair, and bambu for nanopore data?

Salmon is a fast alignment-free quantifier that works against a provided transcriptome and cannot discover new isoforms. Flair is a long-read-specific pipeline that corrects splice junctions, discovers isoforms, and quantifies them. bambu is an R/Bioconductor package that uses the reference annotation to model novel isoforms and quantifies all transcripts in a unified framework. The choice depends on whether your project requires discovery, quantification, or both.

### Can Salmon be used for novel isoform discovery from nanopore data?

No. Salmon quantifies against a provided transcriptome and does not discover new isoforms. If you need to discover novel isoforms, use Flair or bambu. If you use Salmon with an incomplete transcriptome, reads from novel isoforms may be misassigned to known isoforms with similar sequences.

### How does read length affect isoform quantification accuracy?

The LRGASP consortium evaluation found that libraries with longer and more accurate sequences produce more accurate transcripts than those with increased read depth. For quantification accuracy, greater read depth improved results. This means that read length is more important for discovery, while read depth is more important for quantification.

### What should I do if my nanopore data has a high error rate?

High error rates can create false splice junctions and reduce isoform discovery accuracy. Tools like Flair perform explicit error correction, and bambu uses a machine learning approach to filter artifacts. The ESPRESSO study demonstrated that tools using error profiles of individual reads improve splice junction identification. You should also assess whether library preparation can be improved to increase read accuracy.

### Do I need biological replicates for isoform quantification?

Yes. The LRGASP study advises incorporating replicate samples when detecting rare or novel transcripts. Replicates capture biological and technical variation and are required for statistical testing of differential isoform expression. A single nanopore run can identify isoforms but cannot support claims about differential expression without replication.

### How do I choose between Flair and bambu?

Choose Flair if you prefer a modular pipeline with explicit error correction and are comfortable with command-line tools. Choose bambu if you prefer an integrated R/Bioconductor workflow that models novel isoforms and quantifies all transcripts in a single framework. Both tools require a reference genome and annotation, and both perform better with well-annotated genomes.

### What quality controls should I include in a nanopore quantification experiment?

Include spike-in controls with known concentrations to assess quantification accuracy. Include technical replicates to measure technical variation. If you are detecting rare or novel transcripts, incorporate orthogonal data such as short-read RNA-seq or quantitative PCR. The LRGASP study specifically recommends orthogonal data and replicates for reference-free approaches.

### When should I escalate to a specialist for help with nanopore quantification?

Escalate if you have persistent low alignment rates that cannot be resolved by quality filtering, implausible isoform structures that suggest systematic errors, or inconsistent results across replicates that cannot be explained by technical variation. Also consider escalation if you are working with a non-model organism and need to build a custom transcriptome, as this requires specialized expertise in genome annotation.

## Related Bioinformatics Guides

- [RNA-Seq vs qPCR: Validation and Comparison](/knowledge/bioinformatics/rna-seq-vs-qpcr-validation-and-comparison)
- [RNA-Seq Alignment: Choosing the Right Tool and Parameters](/knowledge/bioinformatics/rna-seq-alignment-choosing-the-right-tool-and-parameters)
- [RNA-Seq Alignment Tools: STAR, HISAT2, and Beyond](/knowledge/bioinformatics/rna-seq-alignment-tools-star-hisat2-and-beyond)
- [RNA-Seq Databases: Accessing and Using Public RNA-Seq Data](/knowledge/bioinformatics/rna-seq-databases-accessing-and-using-public-rna-seq-data)
- [RNA-Seq Data Analysis in Galaxy: A User-Friendly Platform](/knowledge/bioinformatics/rna-seq-data-analysis-in-galaxy-a-user-friendly-platform)

## References and Further Reading

- [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/). National Center for Biotechnology Information.
- [EMBL-EBI Training](https://www.ebi.ac.uk/training). European Bioinformatics Institute.
- [Bioconductor](https://bioconductor.org/). Bioconductor Project.
- [Galaxy Training Network](https://training.galaxyproject.org/). Galaxy Project.
- [nf-core Documentation](https://nf-co.re/docs). nf-core.
- [The Carpentries Lessons](https://carpentries.org/lessons). The Carpentries.
- [ESPRESSO: Robust discovery and quantification of transcript isoforms from error-prone long-read RNA-seq data.](https://pubmed.ncbi.nlm.nih.gov/36662851). Science advances, 2023.
- [Systematic assessment of long-read RNA-seq methods for transcript identification and quantification.](https://pubmed.ncbi.nlm.nih.gov/37546854). bioRxiv : the preprint server for biology, 2023.
- [Methodologies for Transcript Profiling Using Long-Read Technologies.](https://pubmed.ncbi.nlm.nih.gov/32733532). Frontiers in genetics, 2020.
- [AsaruSim: a single-cell and spatial RNA-Seq Nanopore long-reads simulation workflow.](https://pubmed.ncbi.nlm.nih.gov/39985444). Bioinformatics (Oxford, England), 2025.
- [Identification of cell barcodes from long-read single-cell RNA-seq with BLAZE.](https://pubmed.ncbi.nlm.nih.gov/37024980). Genome biology, 2023.

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