rna-seq-star-deseq2 completeness audit (2026-08-21)

Upstream: snakemake-workflows/rna-seq-star-deseq2 @ v3.1.1 · Port: oxo-flow-rnaseq-star-deseq2 (live-verified).

Single-DAG snakemake pipeline (24 rules): Ensembl ref download + faidx/ bwa/STAR indices → fastp → STAR align (GeneCounts) → 9 RSeQC rules → count_matrix (technical-rep collapse) → deseq2_init → deseq2 × contrasts (ashr) → gene_2_symbol (biomaRt) → pca → multiqc.

Mode axes (config-driven, orthogonal)

axis	options
read source	local PE / local SE / SRA accession (per unit; SRA PE-only)
trimming	on (fastp_se/pe) / off
strandedness	none/yes/reverse per unit
contrasts	arbitrary count; dict form or R expression list(c(...))
model	default interaction model / explicit formula
batch effects	list → additive model terms + PCA
PCA	activate + labels

Per-sample PE/SE mixing rejected by assert.

Gap tiers vs the port

P0: SRA source mode (get_sra via fasterq-dump); single-end mode (fastp_se + SE chains); biomaRt symbol mapping (4-mirror fallback chain — network dependency); arbitrary R-expression contrasts (port pins declared contrasts? verify scope).

P1: none — MIT, all tools OSS, no paid DBs (Ensembl/BioMart free).

P2: trimming toggle, PCA activate/labels, contrast/batch/model config, strandedness column, fastp per-unit adapters, star index/align passthrough args, ref species/release/build.

Upstream gotchas (do not port): mergeReads.activate is a dead toggle; pca.labels as bare string is a latent bug (char-wise extend); get_bioc_species_name assumes _ in species; test submodule stale.

Gates: Ensembl FTP (pinned release), SRA/ENA for get_sra, biomaRt mirrors, snakemake wrappers fetched at runtime, bwa_index mem 369GB (resource gate — index build on the server class).

Verdict

Compact; the ported default (local PE + trimming + simple contrasts) leaves SRA/SE sources + biomaRt mapping + expression contrasts as the main P0. Port scope (24 rules after normalization) matches the upstream rule count closely — the gap is in the branch axes, not the rule list.

Re-verification (2026-08-23, light group)

Engine: latest main (post-v0.14.1) · Box: tx-ubuntu · Mode: real CLI run, not dry-run · -j 2 --keep-going.

56 succeeded, 48 skipped, 0 failed, exit 0, one round, zero repo fixes. Coverage chain (all real execution): local reference override path (the repo's genome_fasta/annotation_gtf config keys, skipping the Ensembl download) → fastp → STAR align + GeneCounts → RSeQC → count matrix → DESeq2 init/contrast → biomaRt gene2symbol live Ensembl queries, all three steps passed (65.7s / 48.1s / 57.3s) → multiqc. All 48 skips are checkpoint reuse from the earlier round plus condition gates — no unexpected skips.