Quick start

This walks through the whole analysis on the built-in synthetic dataset. Nothing is downloaded, and it takes a few seconds.

1. Make a dataset

splicescope simulate --outdir demo_data

demo_data/ now holds what a real experiment would provide, plus the answer key:

file	contents
sj/*.SJ.out.tab	one STAR junction table per sample, four replicates in each of two groups
annotation.gtf	the reference annotation, with coding sequences
groups.tsv	sample and condition for every sample
genome.fa, genome.fa.fai	an indexed genome whose genes carry real open reading frames
truth.tsv	the events that were injected, so that recall can be measured

2. Run the pipeline

splicescope run \
    --sj-dir demo_data/sj \
    --gtf    demo_data/annotation.gtf \
    --groups demo_data/groups.tsv \
    --genome demo_data/genome.fa \
    --outdir results

Leave out --genome and everything except the consequence step still runs.

3. Read the results

file	one row per
annotation_summary.tsv	junction class: how many junctions are annotated, novel_donor, novel_acceptor, novel_combination or cryptic
differential_splicing.tsv	junction: mean Ψ in each group, ΔΨ, beta-binomial p-value and Benjamini-Hochberg q-value
events.tsv	splicing event (SE, MXE, A5SS, A3SS) with its coordinates
event_differential.tsv	event: event-level percent-spliced-in per group, ΔPSI, p and q
consequence.tsv	cassette exon: frame, premature stop, NMD prediction and consequence class
junction_consequence.tsv	splice-site shift: the same, for novel donors and acceptors
enrichment.tsv	gene set, when --gene-sets is given
cryptic_scores.tsv, model_card.json	junction classifier output, when curated --labels are given
figures/	volcano plots, class summaries and consequence panels

The columns and the statistics behind them are defined in Methods.

4. Or do it from Python

Every step is a function that takes and returns a pandas data frame:

from splicescope import annotate, diff, quantify
from splicescope.simulate import simulate_dataset

ds = simulate_dataset(n_genes=20, n_per_group=6, seed=11)
ann = annotate.annotate_junctions(ds.observed, ds.known)   # junction taxonomy
psi = quantify.compute_psi(ann)                             # splice-site usage
dpsi = diff.differential_splicing(psi, ds.groups)           # ΔΨ, LRT, q-values
print(dpsi.sort_values("qvalue").head())

splicescope.demo.run_demo() runs all of it, protein consequences included, in one call. The tutorial notebook goes through the same analysis step by step, with its outputs, and the API reference lists every public function.

Next steps

• Real data: point --sj-dir, --gtf and --groups at your own files. Use the full GENCODE annotation rather than the basic subset; see the note in the README.
• Many samples or a cluster: the Nextflow pipeline.
• Exploring one event interactively: the dashboard.