Galaxy | Tool Preview

Sailfish (version
Built-ins were indexed using default options
in FASTA format
There is a tradeoff here between the distinctiveness of the k-mers and their robustness to errors. The shorter the k-mers, the more robust they will be to errors in the reads, but the longer the k-mers, the more distinct they will be. We generally recommend using a k-mer size of at least 20.
FASTQ file.
Calculates the aggregated gene-level abundance estimations. This file should be eiher a GTF file or tab-delimited format where each line contains the name of a transcript and the gene to which it belongs separated by a tab.
Larger values will reduce memory usage, but may decrease the fidelity of bias modeling results.
Larger values speed up effective length correction, but may decrease the fidelity of bias modeling results.
When this flag is set, if the intersection of the quasi-mappings for the left and right is empty, then all mappings for the left and all mappings for the right read are reported as orphaned quasi-mappings.
If single end reads are being used for quantification, or there are an insufficient number of uniquely mapping reads when performing paired-end quantification to estimate the empirical fragment length distribution, then use this value to calculate effective lengths.
The standard deviation used in the fragment length distribution for single-end quantification or when an empirical distribution cannot be learned.
Reads mapping to more than this many places won't be considered.
Disables effective length correction when computing the probability that a fragment was generated from a transcript. If this flag is passed in, the fragment length distribution is not taken into account when computing this probability.
Use Variational Bayesian EM algorithm rather than the traditional EM angorithm for optimization
This option will discard orphaned fragments. This only has an effect on paired-end input, but enabling this option will discard, rather than count, any reads where only one of the paired fragments maps to a transcript.
This traditional approach works by convolving the FLD with the characteristic function over each transcript.
When generating the gene-level estimates, use the provided key for aggregating transcripts. The default is the "gene_id" field, but other fields (e.g. "gene_name") might be useful depending on the specifics of the annotation being used. Note: this option only affects aggregation when using a GTF annotation; not an annotation in "simple" format.
All hits are considered "Valid".
Fragments that map in a manner other than what is specified by the expected library type will be discarded, even if there are no mappings that agree with the expected library type.
Allow paired-end reads from the same fragment to "dovetail", such that the ends of the mapped reads can extend past each other.
This is mutually exclusive with Gibbs

What it does

Sailfish is a tool for transcript quantification from RNA-seq data. It requires a set of target transcripts (either from a reference or de-novo assembly) to quantify. All you need to run Sailfish is a fasta file containing your reference transcripts and a (set of) fasta/fastq file(s) containing your reads. Sailfish runs in two phases; indexing and quantification. The indexing step is independent of the reads, and only need to be run one for a particular set of reference transcripts and choice of k (the k-mer size). The quantification step, obviously, is specific to the set of RNA-seq reads and is thus run more frequently.

When the quantification output contains a number of columns: (1) Transcript ID, (2) Transcript Length, (3) Transcripts per Million (TPM) and (4) Estimated number of reads (an estimate of the number of reads drawn from this transcript given the transcript’s relative abundance and length).

The first two columns are self-explanatory, the next four are measures of transcript abundance and the final is a commonly used input for differential expression tools. The Transcripts per Million quantification number is computed as described in [1], and is meant as an estimate of the number of transcripts, per million observed transcripts, originating from each isoform. Its benefit over the F/RPKM measure is that it is independent of the mean expressed transcript length (i.e. if the mean expressed transcript length varies between samples, for example, this alone can affect differential analysis based on the K/RPKM.).

Fragment Library Types

There are numerous library preparation protocols for RNA-seq that result in sequencing reads with different characteristics. For example, reads can be single end (only one side of a fragment is recorded as a read) or paired-end (reads are generated from both ends of a fragment). Further, the sequencing reads themselves may be unstraned or strand-specific. Finally, paired-end protocols will have a specified relative orientation. To characterize the various different typs of sequencing libraries, we've created a miniature "language" that allows for the succinct description of the many different types of possible fragment libraries. For paired-end reads, the possible orientations, along with a graphical description of what they mean, are illustrated below:


The library type string consists of three parts: the relative orientation of the reads, the strandedness of the library, and the directionality of the reads.

The first part of the library string (relative orientation) is only provided if the library is paired-end. The possible options are:

I = inward
O = outward
M = matching

The second part of the read library string specifies whether the protocol is stranded or unstranded; the options are:

S = stranded
U = unstranded

If the protocol is unstranded, then we're done. The final part of the library string specifies the strand from which the read originates in a strand-specific protocol — it is only provided if the library is stranded (i.e. if the library format string is of the form S). The possible values are:

F = read 1 (or single-end read) comes from the forward strand
R = read 1 (or single-end read) comes from the reverse strand

So, for example, if you wanted to specify a fragment library of strand-specific paired-end reads, oriented toward each other, where read 1 comes from the forward strand and read 2 comes from the reverse strand, you would specify -l ISF on the command line. This designates that the library being processed has the type "ISF" meaning, Inward (the relative orientation), Stranted (the protocol is strand-specific), Forward (read 1 comes from the forward strand).

The single end library strings are a bit simpler than their pair-end counter parts, since there is no relative orientation of which to speak. Thus, the only possible library format types for single-end reads are U (for unstranded), SF (for strand-specific reads coming from the forward strand) and SR (for strand-specific reads coming from the reverse strand).

A few more examples of some library format strings and their interpretations are:

IU (an unstranded paired-end library where the reads face each other)
SF (a stranded single-end protocol where the reads come from the forward strand)
OSR (a stranded paired-end protocol where the reads face away from each other,
     read1 comes from reverse strand and read2 comes from the forward strand)


Correspondence to TopHat library types

The popular TopHat RNA-seq read aligner has a different convention for specifying the format of the library. Below is a table that provides the corresponding sailfish/salmon library format string for each of the potential TopHat library types:

TopHat Salmon (and Sailfish)
  Paired-end Single-end
-fr-unstranded -l IU -l U
-fr-firststrand -l ISR -l SR
-fr-secondstrand -l ISF -l SF

The remaining salmon library format strings are not directly expressible in terms of the TopHat library types, and so there is no direct mapping for them.