Skip to content

rnaseq completeness audit (2026-08-21)

Upstream: nf-core/rnaseq @ 3.26.0 · Port: oxo-flow-rnaseq (live-verified).

Mode matrix (upstream — one workflow, parameter cross-product)

axis options default
aligner star_salmon / star_rsem / hisat2 / bowtie2_salmon star_salmon
pseudo-aligner (add-on) salmon / kallisto (null=off) off
trimmer trimgalore / fastp trimgalore
BAM-input mode --skip_alignment + BAM samplesheet columns
UMI --with_umi + --umi_dedup_tool umitools/umicollapse off
rRNA removal --remove_ribo_rna + sortmerna/ribodetector/bowtie2 off
contaminant screening kraken2 / kraken2_bracken / sylph off
QC engine default multi-tool / --use_rustqc default
STAR engine OSS / --use_sentieon_star / --use_parabricks_star OSS
prokaryotic --prokaryotic (+bowtie2_salmon, CDS featureCounts) off

No differential-analysis workflow at this tag (DESeq2 QC = PCA/clustering only). UHR/UMI legacy params removed.

Gap tiers vs the port

P0 (portable): hisat2 branch (extract_splicesites + build + align — no BAM quantification downstream); star_rsem branch (RSEM prepare-reference + calculateexpression + merge + tximport); bowtie2_salmon (+prokaryotic config); pseudo-alignment add-on (salmon/kallisto quant + tximport); BAM-input mode; UMI mode (extract + dedup_umitools/umicollapse + prepareforrsem); rRNA removal (3 tools); contaminant screening (kraken2/bracken/sylph); fastp trimmer variant; stringtie ignore-gtf merge mode.

P1 (license): sentieon (paid license secret), parabricks (commercial NGC container + GPU), SILVA-based SortMeRNA default DB (non-academic license — bowtie2/ribodetector are the license-free routes; document).

P2 (config): all --skip_* QC gates, --rseqc_modules list, --bam_csi_index, strand thresholds, quant libtype args, --save_* publish gates, --stringtie_ignore_gtf, --gencode, --skip_quantification_merge, --min_mapped_reads, extra-tool args, iGenomes.

External deps: iGenomes S3, nf-core/configs fetch, quay.io + Seqera Wave containers, test data HTTPS.

Verdict

Large P0 surface: 2 aligner backends + 2 add-on quantifiers + 4 orthogonal feature modes (UMI/rRNA/contaminant/BAM-input) beyond the ported star_salmon default.

Re-verification (2026-08-23, heavy group 3/3 — heavy 收官)

Engine: latest main (post-v0.14.1) · Box: tx-ubuntu · Mode: real CLI run, not dry-run (checkpoint cleared, forced real execution).

Cumulative 68 rules really executed, 0 real failures — first run 65/15/2 (both failures were env-build: genomeinfodbdata post-link proxy truncation + the deseq2 name collision), then a resume chain: TLS-local conda CA fix → tximport → deseq2 env rebuild (r-optparse added) → deseq2_qc → final 2/80/0 exit 0. Coverage chain (real): fastp → STAR+quantMode → biotype → qualimap → dupradar → bigwig fw/rev/combined → tximport → summarizedexperiment → deseq2_qc → multiqc. All box-side lessons cataloged (stale env-cache / cross-repo name collision → issue #159 / post-link proxy truncation).