RNA-Seq Alignment: Choosing the Right Tool and Parameters

RNA-seq alignment is the computational step where sequencing reads are mapped to a reference genome or transcriptome before expression quantification and differential analysis. The choice of aligner and its parameters directly affects gene expression estimates, isoform detection, and downstream biological conclusions. This article compares STAR, HISAT2, and Salmon for common experimental designs and provides concrete parameter selection guidance based on published benchmarking studies.

What RNA-Seq Alignment Does and Why It Matters

RNA-seq produces millions of short sequencing reads that must be assigned to genomic locations before expression levels can be calculated. Unlike DNA sequencing reads, RNA-seq reads span exon-exon junctions because introns are removed during splicing. Aligners must account for these splice junctions when mapping reads to the genome.

The alignment step is also a technical detail. Published comparisons show that different alignment and quantification pipelines can produce substantially different gene expression values, with consequences for biological interpretation. A systematic evaluation of fifty gene profiling pipelines found that while the best pipelines correlated well with true expression values in simulated data, the error in estimated expression varied considerably across genes, and a small set of genes had consistently high error across datasets and methods. The same study concluded that although mapping software matters, the quantification method makes a greater difference to the results than the aligner itself.

The practical implication is that alignment tool choice should be made deliberately, with awareness of how it interacts with quantification and downstream analysis steps. Researchers should also document their pipeline choices so results can be interpreted and reproduced by others.

Core Principles of RNA-Seq Alignment

Spliced Alignment Versus Unspliced Alignment

Genomic aligners designed for DNA reads, such as BWA and Bowtie, do not natively handle splice junctions. RNA-seq aligners fall into two broad categories. Spliced aligners like STAR and HISAT2 map reads directly to the genome and identify exon-exon junctions during alignment. Transcriptome-based approaches, including Salmon in alignment mode, map reads to a reference transcriptome instead of the genome.

The choice between genome-based and transcriptome-based alignment affects what downstream analyses are possible. Genome-based alignments support detection of novel splice junctions, variant calling, and visualization of reads in genomic context. Transcriptome-based quantification is faster but does not directly reveal genomic features such as unannotated splice sites.

Reference Genome and Annotation Dependencies

Every alignment pipeline depends on the quality and completeness of the reference genome and gene annotation. The reference genome provides the genomic sequence, while the annotation file defines transcript structures. Mismatches between the annotation and the actual transcriptome of the sample, such as missing isoforms or incorrect exon boundaries, will produce alignment artifacts and biased quantification.

Researchers should use the most recent reference genome assembly and matching annotation from authoritative sources. The National Center for Biotechnology Information provides reference genomes and annotation resources for many organisms. The European Bioinformatics Institute offers training materials on using these resources effectively.

Read Length and Sequencing Depth Considerations

Read length and sequencing depth influence aligner performance and parameter choices. Longer reads provide more information for mapping across splice junctions and repetitive regions. Paired-end reads improve mapping confidence because both ends of a fragment must be consistent with the same genomic locus.

Simulation studies have shown that sequencing depth and the relative abundance of different isoforms affect quantification accuracy. When interpreting results for short transcripts, caution is needed because current RNA-seq technologies have limitations in resolving transcript isoforms, including limited read length and the cost of deep sequencing.

At a Glance: Aligner Comparison for Common Use Cases

STAR: Spliced Transcripts Alignment to a Reference

How STAR Works

STAR uses a seed-and-extend strategy with a suffix array index of the reference genome. It finds maximal mappable prefixes for each read, then stitches these seeds together to form the full read alignment. This approach allows STAR to detect splice junctions efficiently because it can join seeds across introns.

STAR is widely used in differential expression pipelines and has been evaluated in multiple benchmarking studies. In a comparison of alignment tools for clinical breast cancer research using formalin-fixed paraffin-embedded samples, STAR generated more precise alignments than HISAT2, particularly for early neoplasia samples. The same study found that HISAT2 was prone to misaligning reads to retrogene genomic loci.

Parameter Selection for STAR

The most important STAR parameters control splice junction detection and multi-mapping behavior. The --outSAMtype parameter determines output format, with BAM recommended for downstream analysis. The --quantMode option can generate gene counts directly during alignment, which is useful for differential expression workflows.

For paired-end data, the --outFilterMismatchNmax parameter controls the maximum number of mismatches allowed. The --alignSJoverhangMin parameter sets the minimum overhang for spliced alignments, which affects sensitivity for detecting short exons. Default values work for many datasets, but researchers should adjust parameters when working with unusual read lengths or challenging samples.

Memory and Compute Requirements

STAR requires substantial memory because it loads the genome index into RAM. For the human genome, 30 GB or more of memory is typically needed. This requirement can be prohibitive for researchers working on standard laptops or small servers. Options include using a smaller genome index with the --genomeSAindexNbases parameter or running STAR on a compute cluster.

The time required for STAR includes a genome index generation step that is performed once per reference genome. After indexing, alignment is fast. The ARPIR pipeline comparison found that the combination of HISAT2 for alignment, featureCounts for quantification, and edgeR for differential analysis had the shortest times and lowest RAM consumption among eight tested pipelines, while STAR-based pipelines were competitive in accuracy.

HISAT2: Hierarchical Indexing for Spliced Alignment

How HISAT2 Works

HISAT2 uses a hierarchical index based on a global genome index and many small local indexes. This design reduces memory usage compared to STAR while still supporting spliced alignment. HISAT2 is the successor to HISAT and TopHat2 and is commonly used in RNA-seq pipelines.

The ARPIR study identified HISAT2 combined with featureCounts and edgeR as the pipeline with the best balance of speed, memory usage, and sensitivity among the eight pipelines tested. This finding supports HISAT2 as a reasonable default choice for standard differential expression experiments where compute resources are limited.

Parameter Selection for HISAT2

HISAT2 parameters include --rna-strandness for specifying library strandness, which is important for accurate read counting. The --dta option instructs HISAT2 to output alignments suitable for transcript assembly with StringTie. The --max-intronlen parameter controls the maximum intron length accepted, which should be set based on the organism being studied.

For paired-end reads, the --fr or --rf options specify the expected fragment orientation. Getting strandness wrong will substantially reduce the number of properly counted reads and bias expression estimates.

Known Limitations

The breast cancer study comparing HISAT2 and STAR found that HISAT2 was prone to misaligning reads to retrogene genomic loci. Retrogenes are processed pseudogenes that lack introns and can attract reads from their parent genes. This misalignment can inflate expression estimates for retrogene loci and reduce sensitivity for detecting differentially expressed genes.

Researchers working with samples where retrotransposition is common, such as cancer tissues, should be aware of this limitation. Cross-checking alignment results with STAR or inspecting specific loci of interest can help identify whether retrogene misalignment is affecting results.

Salmon: Alignment-Free and Alignment-Based Quantification

How Salmon Works

Salmon uses a dual-phase approach. In the quasi-mapping phase, it determines the likely transcriptome origin of each read without full base-to-base alignment. In the quantification phase, it uses an expectation-maximization algorithm to estimate transcript abundances while accounting for biases such as GC content and sequence-specific biases.

The BEERS2 simulation study measured the effect of several settings on the Salmon pseudoalignment algorithm. This work demonstrated that Salmon's performance can be influenced by simulation parameters, including library preparation effects such as polyA selection and ribosomal depletion. Researchers should be aware that Salmon's accuracy depends on the completeness of the transcriptome reference and the correctness of the bias models.

Alignment-Free Versus Alignment-Based Modes

Salmon can run in alignment-free mode using quasi-mapping directly on FASTQ files, or in alignment-based mode where reads are first aligned to the genome or transcriptome and Salmon quantifies from the alignments. The alignment-free mode is substantially faster and requires less memory, making it attractive for large studies with many samples.

The evaluation of isoform quantification tools found that recently developed alignment-free tools are both fast and accurate. However, the accuracy of all methods was mainly influenced by the complexity of gene structures, and caution is needed when interpreting quantification results for short transcripts.

When to Use Salmon

Salmon is well suited for studies focused on gene and transcript expression quantification where genomic alignment is not required. This includes large cohort studies, differential expression analysis, and co-expression network analysis. A study of citrus Huanglongbing tolerance used RNA-seq data from 17 public datasets for weighted gene co-expression network analysis, demonstrating the scale at which transcriptome quantification can be applied.

For studies requiring detection of novel splice junctions, variant calling, or visualization of reads in genomic context, a genome aligner such as STAR or HISAT2 is more appropriate. Salmon can be used in alignment-based mode with a genome aligner to combine the benefits of both approaches.

Practical Workflow for Aligner Selection

Step 1: Define the Experimental Question

The experimental question determines whether genome alignment or transcriptome quantification is needed. If the goal is differential expression of annotated genes and transcripts, Salmon or HISAT2 with featureCounts are efficient choices. If the goal includes discovering novel isoforms, detecting splice variants, or integrating with genomic variant data, STAR or HISAT2 with genome alignment is required.

A study of Graves' disease used integrative RNA-seq analysis to examine gene expression, transcript usage, alternative splicing events, and splicing quantitative trait loci. This type of analysis requires genome alignment to identify splicing events and transcript structures, demonstrating that complex experimental questions demand more detailed alignment approaches.

Step 2: Assess Compute Resources

STAR requires substantial memory, often exceeding 30 GB for mammalian genomes. HISAT2 uses less memory and can run on standard workstations. Salmon in alignment-free mode has the lowest resource requirements and can process many samples quickly.

The ARPIR pipeline comparison provides concrete guidance on resource usage. The HISAT2-featureCounts-edgeR pipeline had the shortest times and lowest RAM consumption among tested pipelines. Researchers with limited compute resources should prioritize this combination or use Salmon for quantification.

Step 3: Consider Sample Type and Quality

Sample type influences aligner choice. The breast cancer study found that STAR generated more precise alignments for formalin-fixed paraffin-embedded samples, which are challenging due to RNA degradation and cross-linking. For degraded RNA samples, aligners that tolerate mismatches and short reads may perform better.

RNA-seq data quality should be assessed before alignment. The RNA-QC-Chain pipeline provides comprehensive quality control including sequencing-quality assessment, trimming, contamination filtering, and alignment statistics reporting. Raw data with quality issues can significantly distort analytical results and lead to erroneous conclusions.

Step 4: Match Aligner to Downstream Analysis

The quantification method makes a greater difference to results than the mapping software. Researchers should select the aligner and quantifier as a coordinated pipeline instead of choosing each tool independently. The interaction between alignment and quantification affects the number of differentially expressed genes detected.

A study of RNA-seq pipeline impact on differentially expressed gene detection found moderate linear correlation between the number of DEGs detected and the percentage of reads aligned with zero mismatch. This finding suggests that alignment stringency parameters influence downstream results and should be chosen deliberately.

Step 5: Document and Validate the Pipeline

Reproducibility requires documenting the exact software versions, parameters, and reference files used. The FAIR Guiding Principles emphasize that research data should be findable, accessible, interoperable, and reusable. Applying these principles to analysis pipelines means recording all computational steps in a form that others can repeat.

Validation can include running multiple aligners on a subset of samples and comparing results. If STAR and HISAT2 produce substantially different expression estimates for specific genes, investigating those genes can reveal whether alignment artifacts are responsible.

Parameters That Matter Most

Mismatch and Indel Tolerances

The number of mismatches allowed affects mapping sensitivity and specificity. Allowing too many mismatches can map reads to incorrect locations, while allowing too few can discard reads with genuine biological variation. The --outFilterMismatchNmax parameter in STAR and the equivalent options in HISAT2 control this tradeoff.

For samples with high genetic diversity or RNA editing, more mismatches may be needed. For well-characterized reference genomes and samples from the same species, default mismatch settings are usually appropriate.

Splice Junction Parameters

Splice junction detection parameters control the minimum and maximum intron lengths accepted and the minimum overhang required for spliced alignments. These parameters should be adjusted based on the organism. For example, mammalian introns can be very long, while fungal introns are typically short.

The --alignSJoverhangMin parameter in STAR sets the minimum overhang for spliced alignments. Lower values increase sensitivity for short exons but may increase false junction calls. The --alignIntronMax parameter sets the maximum intron length, which should be set high enough to capture all genuine introns in the organism.

Multi-Mapping Reads

Reads that map to multiple genomic locations, such as reads from repetitive elements or gene families, present a challenge for alignment. Aligners can report all mappings, report one mapping, or discard multi-mapping reads. The choice affects expression estimates for multi-copy genes.

The --outFilterMultimapNmax parameter in STAR controls the maximum number of loci a read can map to and still be reported. Setting this to 1 reports only uniquely mapping reads, which is conservative but may underestimate expression of multi-copy genes. Higher values include multi-mapping reads but require a strategy for distributing counts among locations.

Strandness

Library strandness describes whether the sequenced reads correspond to the original RNA strand or its complement. Setting the correct strandness parameter is essential for accurate read counting. The --rna-strandness option in HISAT2 and the equivalent options in other tools should match the library preparation protocol.

Getting strandness wrong will cause reads to be counted on the wrong strand, substantially reducing the number of correctly assigned reads and biasing expression estimates. Most library preparation kits specify the strandness of their output, and this information should be recorded in the experimental metadata.

Records and Measurements for Alignment Quality

Alignment Statistics to Track

Every alignment run produces statistics that should be reviewed before proceeding to quantification. Key metrics include the total number of reads, the percentage of reads aligned, the percentage of reads uniquely aligned, and the percentage of reads aligned to multiple loci. Low alignment rates can indicate contamination, reference mismatch, or poor data quality.

The RNA-QC-Chain pipeline reports alignment statistics including read number, alignment coverage, sequencing depth, and paired-end read mapping information. These metrics provide a quality gate before downstream analysis.

Comparing Alignment Rates Across Samples

Alignment rates should be consistent across samples in a study. A sample with substantially lower alignment than others may have contamination, degradation, or a technical problem. Investigating such samples before including them in downstream analysis can prevent spurious results.

The RNA-seq preprocessing study for bladder cancer subtyping found that preprocessing choices significantly influenced molecular subtype classification. This finding underscores that alignment and quantification choices affect also expression values but also higher-level biological classifications.

Recording Parameters for Reproducibility

Each alignment run should record the software version, reference genome version, annotation version, and all non-default parameters. This information should be stored with the analysis results so that the pipeline can be reproduced exactly.

The National Institutes of Health Genomic Data Sharing Policy provides guidance on data sharing expectations for NIH-funded research. While this policy focuses on data sharing, the same principles of transparency and reproducibility apply to analysis methods.

Common Failure Patterns in RNA-Seq Alignment

Low Alignment Rates

Low alignment rates, typically below 70 percent for well-matched samples, indicate a problem. Possible causes include contamination with other species, adapter contamination, poor reference genome quality, or RNA degradation. The RNA-QC-Chain pipeline includes steps for detecting ribosomal RNA contamination and identifying contaminating species.

Strandness Errors

Mis-specified strandness produces a characteristic pattern where reads align to the wrong strand. This error is often detected only when downstream analysis shows unexpected results, such as antisense expression for most genes. Checking strandness on a small subset of reads before running the full pipeline can prevent this problem.

Retrogene Misalignment

The breast cancer study found that HISAT2 was prone to misaligning reads to retrogene genomic loci. This pattern can inflate expression estimates for processed pseudogenes and reduce sensitivity for detecting differentially expressed genes. Cross-checking with STAR or inspecting specific loci can identify this issue.

Memory Exhaustion

STAR's high memory requirements can cause failures on machines with insufficient RAM. The error message typically indicates that the genome index could not be loaded. Solutions include using a smaller index, reducing the number of threads, or moving to a machine with more memory.

Reference Mismatch

Using a reference genome that does not match the sample species or strain produces low alignment rates and biased expression estimates. The reference genome should be chosen to match the sample as closely as possible, and the reference version should be recorded.

Limitations of Alignment Tools and Interpretation

Transcript Isoform Complexity

The evaluation of isoform quantification tools found that the accuracy of all methods was mainly influenced by the complexity of gene structures. Genes with many isoforms or highly similar isoforms are difficult to quantify accurately regardless of the aligner used. Caution is needed when interpreting quantification results for short transcripts.

Sequencing Depth Limitations

The same evaluation found that both sequencing depth and the relative abundance of different isoforms affect quantification accuracy. Lowly expressed isoforms are difficult to quantify reliably, and increasing sequencing depth can improve but not eliminate this limitation.

Sample-Specific Challenges

Formalin-fixed paraffin-embedded samples present specific challenges for alignment due to RNA degradation and cross-linking. The breast cancer study found that STAR generated more precise alignments for these samples, suggesting that aligner choice matters more for challenging sample types.

Preprocessing Effects on Downstream Analysis

The bladder cancer subtyping study found that log-transformation played a crucial role in centroid-based classifiers, while distribution-free algorithms were more robust to preprocessing variations. This finding demonstrates that alignment and preprocessing choices propagate through the entire analysis pipeline and can affect biological classifications.

Safety and Regulatory Context for Data Handling

Data Sharing Policies

Researchers generating RNA-seq data should be aware of data sharing requirements. The National Institutes of Health Genomic Data Sharing Policy describes expectations for sharing genomic data generated with NIH funding. Understanding these requirements before starting a project can prevent compliance issues later.

Data Management for Reproducibility

The FAIR Guiding Principles provide a framework for making research data findable, accessible, interoperable, and reusable. Applying these principles to RNA-seq data includes storing raw sequencing files, alignment files, and analysis scripts in organized, documented formats.

Professional Escalation Criteria

Researchers should seek expert assistance when alignment results are consistently poor, when samples show unexpected patterns, or when downstream analysis produces results that conflict with biological expectations. Bioinformatics support teams, core facilities, or collaborators with computational expertise can help diagnose alignment problems and select appropriate tools.

Frequently Asked Questions

What is the difference between genome alignment and transcriptome alignment?

Genome alignment maps reads to the complete genomic sequence, including introns, and requires spliced alignment to handle exon-exon junctions. Transcriptome alignment maps reads to the set of known transcript sequences, which is simpler and faster but cannot detect novel splice junctions or genomic variants. STAR and HISAT2 perform genome alignment, while Salmon can work directly on transcript sequences.

How much memory does STAR need for human RNA-seq data?

STAR typically requires 30 GB or more of memory for the human genome because it loads the genome index into RAM. The exact requirement depends on the index parameters and the number of threads used. Researchers with limited memory should consider HISAT2 or Salmon, which have substantially lower memory footprints.

Can Salmon replace STAR for differential expression analysis?

Salmon can replace STAR for differential expression analysis when the goal is quantifying annotated genes and transcripts. The evaluation of isoform quantification tools found that alignment-free tools are both fast and accurate. However, Salmon does not provide genomic alignments, so it cannot support analyses requiring read positions in the genome, such as novel isoform discovery or variant calling.

What parameters should I adjust for paired-end RNA-seq data?

For paired-end data, set the fragment orientation parameters correctly, such as --fr or --rf in HISAT2. Adjust the maximum fragment length if your library has unusual insert sizes. For STAR, the --alignIntronMax and --alignSJoverhangMin parameters control splice junction detection and should be set based on the organism.

How do I know if my alignment quality is acceptable?

Review the alignment statistics for each sample, including the percentage of reads aligned and the percentage uniquely aligned. Compare these metrics across samples in the study. Alignment rates below 70 percent for well-matched samples warrant investigation. The RNA-QC-Chain pipeline provides comprehensive quality control metrics for RNA-seq data.

Why do different aligners produce different expression results?

Different aligners use different algorithms for handling splice junctions, multi-mapping reads, and mismatches. The breast cancer study found that HISAT2 and STAR produced different alignment results, with HISAT2 prone to retrogene misalignment. The quantification method also affects results, and the choice of quantifier can make a greater difference than the aligner.

What is the best aligner for formalin-fixed paraffin-embedded samples?

The breast cancer study found that STAR generated more precise alignments for formalin-fixed paraffin-embedded samples, particularly for early neoplasia samples. These samples have degraded RNA and cross-linking artifacts that make alignment challenging. STAR's accurate splice junction detection and precise alignment make it well suited for this sample type.

How should I document my alignment pipeline for reproducibility?

Record the software version, reference genome version, annotation version, and all non-default parameters for each alignment run. Store this information with the analysis results. The FAIR Guiding Principles provide a framework for making research data and methods findable, accessible, interoperable, and reusable.

Related Bioinformatics Guides

• Alternative Splicing Analysis from RNA-Seq Data
• Structural Comparison and Alignment Algorithms for Protein 3D Structures
• Single-Cell RNA-Seq Analysis Pipelines for Veterinary Immunology
• Single-cell RNA-seq Trajectory Inference and Cell Lineage Tracing
• Single-Cell RNA-seq Clustering and Cell-Type Annotation Pipelines

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Evaluation and comparison of computational tools for RNA-seq isoform quantification.. BMC genomics, 2017.
• RNA-Seq gene profiling--a systematic empirical comparison.. PloS one, 2014.
• Aligning the Aligners: Comparison of RNA Sequencing Data Alignment and Gene Expression Quantification Tools for Clinical Breast Cancer Research.. Journal of personalized medicine, 2019.
• BEERS2: RNA-Seq simulation through high fidelity in silico modeling.. Briefings in bioinformatics, 2024.
• ARPIR: automatic RNA-Seq pipelines with interactive report.. BMC bioinformatics, 2020.
• The Impact of RNA-seq Alignment Pipeline on Detection of Differentially Expressed Genes.. ... IEEE Global Conference on Signal and Information Processing. IEEE Global Conference on Signal and Information Processing, 2014.
• BEERS2: RNA-Seq simulation through high fidelity in silico modeling.. bioRxiv : the preprint server for biology, 2023.
• RNA-QC-chain: comprehensive and fast quality control for RNA-Seq data.. BMC genomics, 2018.
• The importance of data transformation in RNA-Seq preprocessing for bladder cancer subtyping.. 2025.
• Exploring the Genetic Networks of HLB Tolerance in Citrus: Insights Across Species and Tissues.. 2025.
• Commentary: a review of technical considerations for planning an RNA-Sequencing experiment.. 2025.
• ExIR enables prioritizing driver and biomarker genes from omics data in a reference free manner.. 2026.
• Machine learning on alignment features for parent-of-origin classification of simulated hybrid RNA-seq.. 2024.
• Integrated RNA-seq and sQTL analysis reveal immune and splicing regulatory features underlying relapse and remission after treatment of Graves' disease.. 2026.
• Performance comparison of five RNA-seq alignment tools Performance comparison of five RNA-seq alignment tools. 2021.
• Performance comparison of five RNA-seq alignment tools. 2013.
• Unbiased comparison of alignment tools for splice junction detection from RNA-Seq data. International Work-Conference on Bioinformatics and Biomedical Engineering, 2013.
• New evaluation methods of read mapping by 17 aligners on simulated and empirical NGS data: an updated comparison of DNA- and RNA-Seq data from Illumina and Ion Torrent technologies. Neural computing & applications (Print), 2021.
• Differential Expression Analysis of RNA-seq Reads: Overview, Taxonomy, and Tools. IEEE ACM Transactions on Computational Biology and Bioinformatics, 2020.
• Comparing bowtie and BWA to align short reads from a RNA-Seq experiment. Advances in Intelligent and Soft Computing, 2012.

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