Nextflow pipeline

nextflow/main.nf is a DSL2 pipeline that runs the same analysis as splicescope run, as processes that Nextflow can schedule on a laptop, a cluster or the cloud. It needs Nextflow 23.04 or newer and splicescope installed in the environment the processes run in.

Demo mode

With no inputs, the pipeline simulates a dataset (genome included) and analyses it:

nextflow run nextflow/main.nf -profile test

Results are published to results/: the simulated inputs in results/data/ and the analysis in results/out/, with the same files splicescope run writes (see Quick start). An execution report and a timeline are written next to them as HTML.

Real data

Give the three inputs, and optionally an indexed genome for the consequence layer. Each is staged on its own, so they can live in different places:

nextflow run nextflow/main.nf \
    --sj_dir  /path/to/SJ_out_tabs \
    --gtf     gencode.v47.annotation.gtf.gz \
    --groups  groups.tsv \
    --genome  GRCh38.primary_assembly.genome.fa \
    --outdir  results

The .fai index must sit next to the FASTA (GRCh38.primary_assembly.genome.fa.fai). Without --genome the consequence step is skipped and everything else runs.

Parameters

parameter	default	meaning
--sj_dir	none	directory of STAR *.SJ.out.tab files; with --gtf and --groups, switches to real-data mode
--gtf	none	reference annotation, plain or gzipped
--groups	none	TSV with columns sample and condition
--genome	none	indexed genome FASTA; enables protein-consequence prediction
--min_reads	10	minimum reads at a splice site in a sample for its Ψ to be computed
--outdir	results	where results are published
--genes, --replicates	12, 6	size of the simulated dataset in demo mode

Profiles

profile	effect
test	a small simulated dataset on one CPU; what CI runs
docker	runs each process in the python:3.11-bookworm container, after installing the splicescope release that matches the pipeline's version from PyPI inside it
conda	runs each process in the conda environment nextflow/environment.yml (Python 3.11 and the matching splicescope release from PyPI)
none	runs each process in your current environment, where splicescope must already be installed

Both the docker and the conda profile need network access to PyPI. With docker, the install is part of each task's script, so it happens inside the container and never touches your own Python environment. The image is the full Debian one rather than -slim because Nextflow needs ps inside the container to record task metrics.

What CI checks

Every push runs the pipeline three ways: demo mode, real-data mode with the inputs in separate directories, and real-data mode without a genome (which must skip the consequence step). See the nextflow job in .github/workflows/ci.yml.