Support single-end and 3' Tag-seq RNA-seq, and designs from authors' tables

[sr320]

Adds what PRJNA913164 (George et al. 2023, 72 single-end QuantSeq 3′ libraries) needs, as a separate code PR per docs/parallel_study_runs.md.

RNA-seq workflow

• --single_end: one FASTQ per sample (fromFilePairs(size: 1)). fastp, Salmon (-r) and the optional viral QC switch to their single-end forms.

• --tag_seq: for 3′ tag libraries, which give about one read per transcript. It changes three steps:

  Step	Standard RNA-seq	With --tag_seq
  fastp	adapter trimming	also --trim_poly_x
  Salmon	--gcBias	--noLengthCorrection, no --gcBias
  DESeq2	from tximport, with length offsets	DESeqDataSetFromMatrix on counts, no length offset; pseudo-bulk sums raw counts

• Both flags are recorded in the workflow manifest's parameters. CI also runs the raw DAG single-end with --tag_seq under -stub-run.

aree fetch-samplesheet

• --design-table / --design-join / --design-source: for deposits whose BioSamples carry no design, only an isolate code. The study commits the authors' table and it is joined on a shared attribute. Every sample must match exactly one row and every row a sample. The table's sha256 and source are written to ena_provenance.json.

Testing

• pytest: 273 passed (3 new design-table tests); ruff: clean.
• Local stub runs pass for both raw DAGs: paired-end with viral QC and an exclusion, and single-end with --tag_seq.
• Real data: used for the full-depth run of all 72 libraries on sr320/AREE#18, with 222/222 tasks plus 3 cached reruns. The run checks out three ways:
  • the manifest records single_end: true, tag_seq: true;
  • Salmon ran -r … --noLengthCorrection, and DESeq2 logged that it used counts without length offsets;
  • fold-change direction agrees with the authors' published DEG lists for 88–97% of their genes, and for 99.3–99.8% of genes significant in both analyses.

Notes for the reviewer

• Footgun, not fixed here: if a study config sets params.rnaseq.comparison_id, it silently beats --comparison_id on the command line. The workflow resolves params.rnaseq?.comparison_id ?: params.comparison_id, so every override run gets filed under the config's id. This bit [study] GEORGE2023_HEATWAVE_PLOIDY (PRJNA913164) #18's first attempt; the manifest's recorded arms caught it. Study configs that run several comparisons should leave rnaseq.comparison_id null. A guard probably belongs in its own PR.
• Possible conflict: this touches modules/rnaseq/emit_manifest.nf (two new parameters keys), as does the in-progress manifest software-versions fix.

Waiting study branch to rebase after merge: study/PRJNA913164/GEORGE2023_HEATWAVE_PLOIDY (sr320/AREE#18).

🤖 Generated with Claude Code