Salmon

What is Salmon?

Salmon is an RNA-seq quantifier for estimating transcript abundance from sequencing reads. It is widely used for transcript-level expression analysis because it combines fast mapping with bias-aware statistical inference, producing normalized abundance estimates such as TPM alongside estimated fragment counts.

Salmon was designed for annotated transcriptomes rather than de novo assembly. In practice, that means the quality of the reference transcript FASTA strongly influences the quality of the quantification.

How to use Salmon online

ProteinIQ runs Salmon in the browser through a cloud workflow, so transcriptome indexing and quantification can be performed without installing command-line software locally. The online form accepts an uploaded transcript FASTA reference together with either single-end or paired-end FASTQ reads, then returns the main quant.sf table and Salmon metadata files.

Inputs

Settings

Output

Salmon returns the primary abundance table as well as run metadata that records how the job was executed.

The data table shown in ProteinIQ exposes the main quant.sf columns:

How does Salmon work?

Salmon builds an index over the supplied transcriptome, identifies candidate transcript origins for each read or read pair, and then estimates abundances with a probabilistic inference procedure. The 2017 Salmon paper describes this as a dual-phase approach: an online phase learns experiment-specific parameters while processing fragments, followed by an offline optimization step that refines transcript abundance estimates.

Selective alignment adds an alignment-scoring stage on top of lightweight mapping. This reduces false assignments that can occur when reads match multiple similar transcript sequences or resemble unannotated genomic regions. In current Salmon workflows, selective alignment is often paired with decoy-aware references for improved specificity; on ProteinIQ, the exposed Validate mappings option enables the selective-alignment validation step for uploaded transcriptomes.

Bias correction is central to Salmon's design. Sequence-specific effects, fragment-level GC bias, and effective transcript length all influence how raw fragment evidence is translated into expression estimates. These corrections are why Salmon output should be interpreted as model-based abundance estimates rather than simple read counts.

Interpreting results

TPM is useful for comparing transcript abundance within a sample because it normalizes for both transcript length and sequencing depth. A higher TPM indicates that a larger share of the sequenced RNA is attributed to that transcript, but TPM values are still relative and should not be treated as absolute molecule counts.

Estimated Reads is closer to an assigned fragment count, but it is also model-derived because ambiguously mapping reads are distributed probabilistically. For transcript families with extensive sequence overlap, the distinction between TPM and Estimated Reads is less important than the underlying identifiability of the transcripts in the reference.

Effective Length matters when short transcripts or libraries with different fragment distributions are compared. If two transcripts have similar raw support but different effective lengths, the shorter effective transcript can receive a higher normalized abundance estimate.

Limitations

• Salmon can quantify only transcripts present in the uploaded reference FASTA. Missing isoforms, truncated models, or redundant transcript records can distort abundance estimates.
• Transcript-level quantification remains difficult when isoforms share most of their sequence. In those cases, abundance may be spread across several similar transcripts.
• Library-type inference is convenient but not infallible. For stranded RNA-seq experiments, known library orientation is usually preferable to automatic detection.
• Bootstrap replicates improve uncertainty assessment but increase runtime and output size.
• Salmon quantifies against a transcriptome reference. It does not replace splice-aware genome alignment when the goal is novel transcript discovery, splice junction analysis, or variant-aware read inspection.

Related tools

• FASTQ to FASTA converter: Converts sequencing reads into FASTA when downstream workflows require sequence-only input.
• FASTA splitter: Breaks large transcriptome FASTA files into smaller reference sets for targeted analyses.
• TXT to FASTA converter: Converts plain transcript lists or exported sequence text into FASTA before building a Salmon reference.