Bio Pipeline Runner

When to use

Use this when the user wants to run a standardized nf-core pipeline on sequencing data and needs the samplesheet, config, and command assembled correctly the first time — rnaseq (expression), sarek (variants), or atacseq (open chromatin).

Not for: custom one-off analyses with no nf-core equivalent, QC of an already-produced count matrix (use bio-single-cell-qc), or interpreting biology from finished results.

Method

1. Match pipeline to the assay. Decision: rnaseq for DGE/quantification, sarek for germline/somatic variants, atacseq for open chromatin. If the assay doesn't map cleanly, stop and clarify rather than force-fit.
2. Pin the version with -r <release> and pick the reference (iGenomes key or explicit FASTA/GTF).
3. Build the samplesheet in the pipeline's exact schema:
   • rnaseq: sample,fastq_1,fastq_2,strandedness
   • sarek: patient,sample,lane,fastq_1,fastq_2
   • atacseq: sample,fastq_1,fastq_2,replicate Validate paths exist and paired-end mates match.
4. Choose the execution profile: docker/singularity/conda for containers plus an executor (local/slurm) and resource limits via -c. Decision: singularity on shared HPC (no root), docker on a personal box.
5. Dry-run with -profile test or -stub to catch schema errors before burning compute.
6. Launch, then inspect MultiQC and pipeline_info execution reports for failures and resource waste.

Example

Bulk RNA-seq, 6 samples, reverse-stranded, on a SLURM cluster. Choose nf-core/rnaseq -r 3.14.0, genome GRCh38. Samplesheet rows read ctrl_1,/data/c1_R1.fq.gz,/data/c1_R2.fq.gz,reverse. Profile: -profile singularity with a slurm.config capping 16 CPU / 72 GB per process. -profile test passes, then the full run. MultiQC STAR alignment rate ~92% confirms the stranded setting was right.

Pitfalls

• Leaving strandedness as auto when the kit is known — silently wrong DGE if guessed incorrectly.
• Not pinning -r, so a later default-branch change makes the run irreproducible.
• Skipping the -profile test dry-run and discovering a samplesheet typo after hours of alignment.
• Requesting more memory per process than the node has, so every job pends forever.

Output format

# nf-core Run Plan — <pipeline>
Pipeline: nf-core/<name>  Version: -r <release>  Genome: <key/FASTA+GTF>  Rationale: <one line>

## samplesheet.csv
<header + rows in exact schema>

## Command
nextflow run nf-core/<pipeline> -r <ver> \
  --input samplesheet.csv --outdir results \
  --genome <key> -profile <container>,<executor> -c <custom.config>

## Post-run checklist
MultiQC sections to check | common failure modes | where results land

The run executes on the user's compute — report commands, never fabricate results.

Reference

Pipeline selection

nf-core/rnaseq (v3.14+) key params

• --input samplesheet: sample,fastq_1,fastq_2,strandedness — strandedness ∈ forward (e.g. some Ligation kits), reverse (Illumina TruSeq stranded / dUTP — most common), unstranded, auto. Set it explicitly when the kit is known; auto (salmon inference) is a fallback, not a default.
• Aligner: --aligner star_salmon (default, recommended — gives gene + transcript quant), star_rsem, or --pseudo_aligner salmon/kallisto (alignment-free, fast).
• Reference: --genome GRCh38 (iGenomes) or explicit --fasta/--gtf (iGenomes GTFs are dated — a current GENCODE/Ensembl GTF is often better).
• Useful: --extra_salmon_quant_args, --skip_biotype_qc, --remove_ribo_rna (+ --ribo_database_manifest), --umitools_* for UMI protocols.
• QC bundled: FastQC, trimming (Trim Galore/fastp), STAR logs, RSeQC, Qualimap, dupRadar, Preseq, DESeq2 sample-similarity PCA, all rolled into MultiQC. Sanity check: STAR uniquely-mapped ≥ ~80–90%, assigned reads high, and the strandedness bar plot matches your declared setting.

nf-core/sarek (v3.4+) key params

• --input: patient,sample,lane,fastq_1,fastq_2 (or bam/cram + .crai). Group tumor/normal by shared patient, set status (0=normal,1=tumor) for somatic.
• --wes for exome/panel (add --intervals target.bed — critical for runtime & correctness); omit for WGS.
• --tools: callers e.g. strelka,mutect2,manta,cnvkit,ascat,deepvariant,haplotypecaller,freebayes,vep,snpeff. Germline → haplotypecaller/deepvariant; somatic → mutect2/strelka; SV → manta; CNV → cnvkit/ascat.
• --genome GATK.GRCh38 (bundles known-sites for BQSR); --joint_germline for cohort GVCF joint calling. --step can resume at mapping|markduplicates|prepare_recalibration|recalibrate|variant_calling|annotate.

nf-core/atacseq (v2.1+) key params

• --input: sample,fastq_1,fastq_2,replicate (replicate integer groups technical/biological reps for consensus peaks).
• --read_length (drives MACS2 effective genome size), --narrow_peak (default; omit → broad), --macs_gsize (or auto from genome). Blacklist filtering is applied for supported genomes.
• QC: TSS enrichment, FRiP, fragment-length periodicity (nucleosome laddering), library complexity — inspect in MultiQC.

Reproducibility & profiles

• Always pin -r <release> (e.g. -r 3.14.0); an unpinned run tracks the default branch and drifts.
• Container/env profile: -profile docker (personal box, root), singularity/apptainer (shared HPC, no root — most common), conda (last resort), plus executor via a -c custom.config (process.executor = 'slurm', queue, withName/withLabel resource overrides). nf-core resource labels: process_low (2 CPU/12 GB), process_medium (6 CPU/36 GB), process_high (12 CPU/72 GB), process_high_memory (200 GB, e.g. STAR index for large genomes). Never request more than a node offers or jobs pend forever.
• Rough guidance: human WGS sarek ≈ 100–300 GB scratch + tens of core-hours/sample; STAR RNA index needs ~30–40 GB RAM; use -resume to reuse cached tasks; set -work-dir on fast scratch and clean it after.
• Dry run first: -profile test,<container> (tiny bundled data) or -stub catches samplesheet/schema errors before burning real compute. --outdir is required; results land under it with pipeline_info/ (execution_report.html, timeline, trace) and multiqc/.