Skip to content

v2.3.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 02 Jul 04:21
v2.3.0
5deb7aa

salmon 2.3.0 adds genome-alignment quantification, a deterministic alignment mode, and a dependency-graph cleanup. It is fully backward compatible with 2.2.x (CLI, index, and output formats unchanged; RAD reading is back-compatible).

Highlights

Genome-alignment quantification (-a <genome.bam> --annotation <gtf>)

Quantify directly from a genome-aligned, name-grouped BAM (STAR, HISAT2, …) plus an annotation — no second transcriptome alignment. Spliced alignments are projected into transcriptome coordinates with bramble, then quantified through the same RAD/EM/output stack as every other mode.

  • --genome <fasta> reconstructs transcript sequences so --seqBias/--gcBias/--posBias work; --juncMissDiscount <f> penalizes unannotated junctions.
  • Inherently deterministic (byte-identical across thread counts); composes with --numBootstraps/--numGibbsSamples, -g/--geneMap, and -l A.
  • Short-read (Illumina) focused; for long reads use oarfish.
  • On simulated data it tracks direct transcriptomic quantification, matching a reference genome→transcriptome projection (e.g. STAR --quantMode TranscriptomeSAM); the residual read-vs-genome gap is inherent to genome alignment (paralog-locus collapse), not the projection.

See the Genome-alignment quantification guide.

Deterministic alignment mode (-a <txome.bam> --deterministic)

Byte-identical quant.sf across thread counts for transcriptome-aligned BAMs. Placements are scored by the BAM AS tag by default (a single BAM pass).

  • --errorModel (opt-in, needs -t) trains an order-independent alignment error model — integer transition counts merged across threads, normalized once — restoring the classic error-modeled weighting while staying reproducible (a second BAM pass).
  • Why AS by default: across every truth-bearing benchmark (uniform and realistic position-dependent Illumina errors, 50 bp and 76 bp), AS scoring is at least as accurate as an error model while being faster. --errorModel is there for parity with salmon's classic behavior, or aligners that emit no usable AS.

See the deterministic alignment guide.

Dependency-graph unification

noodles (and its sub-crates) and ksw2rs now each resolve to a single version across salmon and its COMBINE-lab dependencies (bramble-rs 0.1.5, libradicl 0.14.2), shrinking the build graph. No user-facing change; the online alignment path is byte-for-byte consistent with 2.2.1.

Compatibility

  • CLI, index format, and quant.sf/quant.genes.sf/inferential-replicate outputs are unchanged from 2.2.x.
  • The RAD schema gains a backward-compatible score_kind tag on selective-alignment RAD files; older RAD (salmon or piscem) reads unchanged.

Install salmon-cli 2.3.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/COMBINE-lab/salmon/releases/download/v2.3.0/salmon-cli-installer.sh | sh

Download salmon-cli 2.3.0

File Platform Checksum
salmon-cli-aarch64-apple-darwin.tar.xz Apple Silicon macOS checksum
salmon-cli-x86_64-apple-darwin.tar.xz Intel macOS checksum
salmon-cli-aarch64-unknown-linux-gnu.tar.xz ARM64 Linux checksum
salmon-cli-x86_64-unknown-linux-gnu.tar.xz x64 Linux checksum
Assets 15
Loading